Artificial Intelligence: A Modern Approach

Text books are usually cryptic and boring, but this one's actually quite fun to read. It's so easy and fun that I'm actually excited when the professor assigns a new reading. In fact, I liked it so much that I looked up the author to find more of his books, and guess what? The author works at Google. Someone in Google making a user-friendly textbook? I'm in!

Not as good as I remember the first edition being. It seemed like in the past there were more fundamental concepts and real techiniques to use while this one seems very theory centric. Good book for covering all the bases however and not too bad a read.

Having bought this for the online AI class I didn't notice the device limit. In the day and age that people have several devices capable of reading Kindle books (I have 7 including work and home computers), there is no excuse for limiting it to two devices. I understand the concern with people sharing the books, but when will companies learn treating your consumers like criminals is a losing proposition.

This book was purchased for a college graduate class (Advanced Artificial Intelligence). It is a good book. The sections I have read have been clear and understandable. In all this is one of the better textbooks I have used.

I loaded the Kindle version up on my iPad. Great. But I needed it on my laptop (Mac) for class. I tried to load it with Kindle for Mac and it claims it has exceeded the maximum allowed licenses. I am extremely pissed. I bought the hard back for a small fortune and only slightly smaller fortune for e-access and I can't even read it on a measly two devices? This is totally and completely unacceptable.

This review is for the kindle version, not the content. For this much money I'd really expect a better conversion. It is really just a pdf, not a real ebook. You are not able to change font sizes o background colors, they basically did the absolute bare minimum.

the store is usually pretty good about books. This one in particular I wound up getting brand new for the same price I was gonna buy a used one at my local bookstore. So, definitely check out the store and compare prices

- With AIMA 1st Edition, I had relearned AI anew from a fresh, insightful and wonderfully pedagogical perspective.
Best computer science textbook ever.
- With AIMA 2nd Edition, I got a lot of recent advances in AI brought to me in the same way, even if presented at times in a way that was too concise for a textbook, and read more like an encyclopedia.
Yet, great 2nd Edition.

- This 3rd Edition is alas AIMA 2.1 and not the AIMA 3.0 that I was waiting for. The new material and new insightful way to organize past material are both scant. Certainly not worth the price for those who own the 2nd Edition.

Don't get me wrong, if you are about to buy your first AI textbook, this is a great buy as it is still light years ahead of the competition. But some chapters that were getting really thin and outdated in 2009 did not get significant updating.

This is particularly true for knowledge representation. Missing are all the recent yet already consolidated advances brought about by the new solutions to the frame problem (such as the fluent calculus), default reasoning, abduction-based and case-based diagnosis, rule-based reasoning (such as constraint handling rules, answer sets, object-oriented logic programming etc.), in short, all forms of reasoning that are neither pure deduction, nor probabilistic. Advances on multi-agent reasoning are also not covered. I understand that to summarize AI in 1000 pages many important topics will not make the cut, but I feel, as a researcher on the topic for the past 25 years and lecturer on it for the past 15 years, that this 3rd edition contains obsolete stuff from the 80s (like frames, semantic networks, production systems, situation calculus, etc.) instead of their modern substitute listed above.

In short, after two Herculean efforts, it seems like the authors put far less work in this one. As a result, we are left without an truly comprehensive and up-to-date text to teach AI and agents. I hope the incoming text by David Poole will cover some of the weaknesses of this AIMA 2.1.

This is a reasonable overview of AI - and an amazing achievement to have so much material in one book - but it is increasingly out of date. A lot of the techniques described at length could be fairly described as "Good Old Fashioned AI" and could have been shortened to make way for more powerful modern techniques. Other reviews have given specific details, but machine learning techniques in particular deserve more than one chapter. There is no mention in the index of "bias/variance tradeoffs", an important topic in which good progress has been made lately.

The changes in the third edition mostly amount to shuffling things around a bit. Only one chapter (Chapter 20) was substantially changed. Given the high price of the new edition it is probably not worth the money if you have an older edition. You would be better off to search out a specialised text or material on the web on the new techniques.

Reading the book superficially, it is quite informative and enjoyable. The reference list is very good. However I found when I tried to use and implement the algorithms described I ran into problems. Concretely:

* The pseudo-code is a strange mix of mathematical notation, Python-like code and prose. I found it very hard to turn it into real code, though I did succeed eventually in some cases. Apart from the undefined nature of the 'language', the variable names and function names chosen are often very uninformative and terse. You might have P (in bold) as one variable, and other p (in italics). Variable names like "var" and "value" abound. The pseudo-code does not follow the conventions described in the appendix. You need to have a high tolerance for frustration.

* The writing style is terse and mathematical. New notations are introduced freely and used hundreds of pages later without explanation e.g. the use of alpha as a normalizing factor in Bayesian calculations. There is no glossary of symbols that you can refer to. It is necessary to undertake a tedious search of the previous sections of the book, hoping for enlightenment.

Also there is much use of phrases such as "we therefore see". Often it is very unclear how we do "see" that the conclusion is true. Again the reason may be something that was covered several chapters ago (or in at least one case, in later pages). Perhaps this reflects the terse mathematical approach where you are presumed to have memorized the prior portions of the text, it is assumed you are used to absorbing new notations at a high rate, and that the greatest sin of all is to repeat yourself or to state the obvious.

My suggestion would be to borrow this book from a friend or a library to get an overview of Good Old Fashioned AI. Then read some course notes on machine learning (eg Stanford CS229) to get an update on machine learning. Then purchase specialized texts for areas you actually want to use. A lot of good material is legitimately available online eg Sutton's book on Reinforcement Learning but you are going to have to buy some books.

Did you know that this book is available for $125 less, as a version published for Indian students?

Yep, for less than $25, you can have the EXACT SAME CONTENT, in a less expensive paperback form.

Yet, strangely, this version has a large red warning on the front cover, saying having this book outside of India is "UNAUTHORIZED"... which means... nothing. It's not illegal (per U.S. court), but Pearson would sure like you to THINK it is. Why? So they can continue to charge American kids prices that border on usury, hand-in-hand with professors that demand their kids buy these overpriced editions, even if there are barely any changes from one edition to the next... besides the price, of course.

Other countries don't LET publishers do this to THEIR kids... don't participate in it for OUR kids. Buy the Indian version, hit Pearson the only place it will hurt them: their bottom line. Maybe if enough people do it, they'll change.

Sadly the third eddition feels like part of the textbook racket where new editions are created with more of an aim at increasing profits than increasing quality. Besides that gripe I found the book to be a better than average textbook. I've used it much more heaviled that many other textbooks I've purchased.

Being the classic AI textbook notwithstanding, I did not like it. It is too wordy, dry, and complex. For me, I prefer "Contemporary AI" by Richard Neapolitan. It is more reader friendly although less comprehensive.

Russell and Norvig's AI: A Modern Approach is THE best AI text out there. At 932 pages it is encyclopedic, it has nearly everything. So what is missing? How could it be improved? Probably the worst thing about the book is the binding. I am not sure that you can read it from cover to cover without some pages coming loose. Perhaps its the length. Perhaps it needs to be split into two volumes. I am not a fan of pseudocode and all the algorithms are in pseudocode. I think the right compromise between detailed practical code and tutorial compactness is something like the code in Jackson's text Expert Systems. I realize this might make a long book even longer but I still think some examples in Lisp, Prolog, etc. would be an improvement. There are a few things missing. Some detail on case-based reasoning is needed and some newer topics like hybrid systems and rough sets. Also, more on parallel computing for AI. Occasionally I was annoyed by the references. On page 27 the authors attribute a story to Heckerman's 1991 thesis. The thesis contains no such story. The reference should have been to a private communication. By now you might think I hate the book. No. I am suggesting improvements. I repeat. It is THE BEST SINGLE AI TEXT IN PRINT. But you will not be able to teach the whole book in a single AI course. Not even a two semester course.

I have worked with this book during two courses I have had on AI, and I must say that this is definitely one of the best textbooks I have read in the field of computer science and algorithms. The book thoroughly covers subjects from search algorithms, reducing problems to search problems, working with logic, planning, and more advanced topics in AI such as reasoning with partial observability, machine learning and language processing. I have not yet had time to study the more advanced topics, but I can say that the first half of the book dealing with searching, logic and planning are very well written and understandable by most students who know basic programming. Algorithms and data structures are mostly introduced along the way, but some prior knowledge, such as knowing the basics of graph theory etc., is probably an advantage.

The book is mostly written in a concise and easily digestible language, but some sections could probably have been written in fewer words.

Overall, this book is one of my favorite textbooks!

AI: A Modern Approach is a great introduction to a good range of topics in the field of AI. Going into this book, I knew nothing of AI. The first few chapters cover intelligent agents, searching, and various search algorithms such as the basics like Depth-First and Breadth-First, and then the book introduces some more intelligent algorithms like A*, SMA*, Iterative Deepening, and a few more. Other topics included in the book are planning, logic (if you're new to logic, I might recommend some supplimentary material; it's very important to understand everything if you're interested in AI). I've read several introductory books on AI, and I would definately rate this one as the best!

As a physics student, I have been particularly interested in Bayesian Models of Cognition for some time, but I have had trouble finding the appropriate bibliography to build a strong knowledge base before going into more advance topics. I found this book because it is one of Tom Griffiths' recommended readings in Bayesian methods. I started reading in part IV: Uncertain Knowledge and Reasoning, and then I went through the entire book.
This book was that strong knowledge base I was looking for, and so much more! It covers, in a very understandable way, most of the areas of interest in Probabilistic Cognition: graphical models, exact inference in Bayesian Networks and the need for sampling methods, Monte Carlo sampling, learning probabilistic models, the EM algorithm, Hidden Markov Models and DBNs, Decision Theory, etc.
The book is very clear in the math, the algorithms and its examples. After reading the book intensively (working out the examples numerically) and doing most of the exercises; I found my self understanding advanced papers on this subjects in a very natural way.

Briefly, if you are interested in Probabilistics models of Cognition, this is the place to start.

As a physics student, I have been particularly interested in Bayesian Models of Cognition for some time, but I have had trouble finding the appropriate bibliography to build a strong knowledge base before going into more advance topics. I found this book because it is one of Tom Griffiths' recommended readings in Bayesian methods. I started reading in part IV: Uncertain Knowledge and Reasoning, and then I went through the entire book.
This book was that strong knowledge base I was looking for, and so much more! It covers, in a very understandable way, most of the areas of interest in Probabilistic Cognition: graphical models, exact inference in Bayesian Networks and the need for sampling methods, Monte Carlo sampling, learning probabilistic models, the EM algorithm, Hidden Markov Models and DBNs, Decision Theory, etc.
The book is very clear in the math, the algorithms and its examples. After reading the book intensively (working out the examples numerically) and doing most of the exercises; I found my self understanding advanced papers on this subjects in a very natural way.

Briefly, if you are interested in Probabilistics models of Cognition, this is the place to start.

As textbooks go, this one is well-organized and illustrated, and in addition to educating you with regards to artificial intelligence, provides a decent background in introductory algorithms and probability.

What? What's that?

He's on his way to city hall?

He's hacked all the security cameras? Who gave him the capabilities to interface with such systems?!

...I'm terribly sorry. I must attend to one of my more unruly projects. Good day.

This is the most complete and comprehensive book I read on a subject of Artificial Intelligence so far and it's very well written as well. If you plan diving into AI really seriously and you are keen to invest some good amount of time going through 1000 pages of this book then I really recommend it for you. Great addition to this book is A.I. course [...] led by coauthor of this book, Peter Norvig and Sebastian Thrun, a Professor of Computer Science and Electrical Engineering at Stanford University. Last three months I spent every day with both this book and A.I. course and it was the most fascinating learning experience I've ever head.

This book covers an amazing array of AI topics. Few books cover predicate logic as well as neural networks and other stoicastic processes. The coverage is usually quite in depth yet still manages to be readable. The bibliography section will be helpful for further reading. It is not for cover to cover reading due to the nature of the subject but could be done with a bit of determination. The rewards for doing so would be well worth it.

An essential core text for anyone interested in artificial intelligence with a particular emphasis on the practical use and implementation of intelligent agents. Thorough and comprehensive, yet highly suitable as an entry-level text and written with a light touch that actually had me chuckling out loud at a couple of points (believe it or not)!

The publication of this textbook was a major step forward, not only for the teaching of AI, but for the unified view of the field that this book introduces. Even for experts in the field, there are important insights in almost every chapter. I recommend it to anyone who wants to have an introductory overview of the state of AI. And I recommend it to experts in the field, who will enjoy its unified description of the field. I especially enjoyed the introductory chapter and the chapter on philosophical issues. I have taught from this book three times now, and it has improved my AI class hugely.

I found this book to be an extremely interesting read. It gave a good overall picture of the developing field of AI as well as it's history. It's a great stepping stone for people who are familiar with discrete math & logic to get an idea of what is on the horizon, but it is not the end-all-be-all of intelligent agent design; in an area of science that changes day by day, I think the authors took a brilliant snapshot and laid a good foundation for future learning. I must admit, however, if I had needed to read this book for a course in school, I probably wouldn't have enjoyed it half as much. But isn't that the nature of school?

Update Jan 2013
--------------
The new Kindle version '(3/e) [Print Replica]' fixes all the previous typesetting problems which were in the original '(3/e)' version. Make sure you get this new version.

Unfortunately for the owners of '(3/e)', the store has classified the update as a different product, even though the content is identical; currently no upgrade is available.

The new version gets 5 stars.

Original Review
-------------
This is not a review of the book contents, but of the Kindle format for this book.

1. I bought the Kindle version to read on both my iPhone (primarily) and computer. Unfortunately this book is not currently formatted for use on iPhone, but it is for other others. I now know to check the "available for these devices" section. Hopefully an iPhone format will be made available.

2. In the printed text, what are probably helpful concept tags in margin are in the Kindle version a complete nuisance. These are formatted in-line (not in the margin), making for some confusing reading and rendering the tags completely ineffective. It would be better to strip them altogether, especially since these terms are already in bold-face in the text. For example:

<quote>
3.3 SEARCHING FOR SOLUTIONS
Having formulated some problems, we now need to solve them. A solution is an action sequence, so search algorithms work by considering various possible action sequences. The possible action sequences starting at the initial state form a search tree with the initial state at the root; the branches are actions and the nodes correspond to states in the state space of the problem. Figure 3.6 shows the first few steps in growing the search tree for finding a route from Arad to Bucharest. The root node of the tree corresponds to the initial state, In(Arad). The first step is to test whether this is a goal state. (Clearly it is not, but it is important to check so that we can solve trick problems like "starting in Arad, get to Arad.") Then we need to consider taking various actions. We do this by expanding the current state; that is, applying each legal action to the current state, thereby generating a new set of states. In this case, we add three branches from the parent node In(Arad) leading to three new child nodes: In(Sibiu), In(Timisoara), and In(Zerind). Now we must choose which of these three possibilities to consider further.
______________
SEARCH TREE
______________
NODE
______________
EXPANDING
______________
GENERATING
______________
PARENT NODE
______________
CHILD NODE
______________
</quote>

You get the picture. In sections with many new terms this nonsense can consume half the page.

3. Same problem as #2, but with the 'pointed finger' symbol to draw attention to an important point. Having the symbol show up between paragraphs, rather than in the margin pointing at the actual text is useless. Besides the text is already italicized.

4. Text underneath figures needs a line break before continuing with the normal text, as it blends in.

No sure if the store or the publisher does the electronic conversion, but please fix these! Clearly an editor or better conversion AI is required. As much as I like electronic books, I would have gone with the printed copy for this one given the current state.

Update 2/23/2012:
-----------------
There is now an iOS compatible format, but the same formatting problems remain.

I bought this for Norvig's class at Stanford. This is actually the paperback international edition which is identical to the hardcover US edition. The textbook is very informative delving into a lot of topics. In particular, it combines some relatively accessible descriptions with a bit more structured pseudo-code (or pseudo-process) information in the style of Introduction to Algorithms.

Thanks to all those who reviewed the first edition.
If you are reading this, you will probably want the
second edition instead. It was published Dec 20, 2002.
Every chapter has been extensively rewritten.
Significant new material has been introduced to cover
areas such as constraint satisfaction, fast propositional
inference, planning graphs, internet agents, exact
probabilistic inference, Markov Chain Monte Carlo
techniques, Kalman filters, ensemble learning methods,
statistical learning, probabilistic natural language
models, probabilistic robotics, and ethical aspects of AI.
For more information, see aima.cs.berkeley.edu

With a light-heartened style, these authors take the reader from the basics to a very good level of knowledge. The reader chooses what level to get to, by means of solving the suggested exercises. Covers the main themes. Obviously aimed to inexperienced readers.

The entire class has had difficulties with the problems in each chapter. It is difficult to determine what they want as an answer, the questions can be answered at many different levels and can get messy. Overall the biggest improvement they could make to this book would be the addition of more examples. The only saving grace I've seen so far in this book has been the interesting bits of history thrown in here and there. If you are a student and the book is required there isn't much you can do, but if not there has to be a better book out there.

...I can only relate my experience from using this as a textbook for an introductory AI class with assigned topics. I think that the choice of material covered in this book is good, but I don't really think it's very clear in most places and there are not enough (or any) examples in most chapters. I have also used Winston's textbook for a different class, and felt that it was much clearer, as well as provided better relevant examples and problems. After having used both, I prefer Winston's text (ISBN 0201533774) to this one.