I'm currently teaching AI. Since it's the standard textbook for AI courses, I decided to use Russel&Norvig's book, and I am really disappointed.

The book is too superficial, trying to cover too much, and their notation and explanations are not always clear. For example, try to understand the Viterbi algorithm for HMMs. It's perfectly clear if you read an introductory article, but this book gives a very confusing idea of how it works. In several other parts of the book the same thing happens.
More often than not I have given other texts to my students.

I do not think using "one big book" is the right approach for teaching AI, because "AI" is too large. If you are teaching undergrad students in a "BS in AI" then you should use specific and in-depth books for each course: knowledge representation, vision, uncertainty, etc.
But if you are (as I am) teaching a short AI course in a Computer Science context, then I think you should probably pick very few subjects and treat them *in depth* -- otherwise your students will have no benefit in taking your course (whatever you tell them in that short time, they could learn by other means).

I enjoyed this book as a student taking an AI class. However, it was too heavy to carry it to the class. I did like book website and Google code page.

On the negative part I'd say the layout of the examples/pseudo code was ...rather inconvenient. LISP style made it a little bit awkward for a person who never saw LISP before. Some examples about evil king and his brother, and such were a little bit off... I'd rather get some real life examples. Problems at the end of chapters did not encourage going and doing it on your own.

I am not sure I'd be able to use this book as a self-study guide, but in the class it did make sense.

Following the accolades in the reviews and having a keen interest in AI (as a physician and computer scientist) - I have dived into this book. It took me more than half a year of stubbornly trying to read and understand it. What a disappointment...
On one hand, the math is inaccessible, least you have a major in computer sciences / statistics, math - or all of the above. It seems some, if not all of the math "proofs" are unnecessary for the matter at hand. Unless there are some sinister motives behind these superfluous math complications - such as providing professors with ammunition for students testing. But why should someone interested in AI - get bogged down in this? Is it really what the authors had in mind?
On the other hand there are not enough examples to follow and the examples that are there - are inconsistent and insufficient (for example: the `wumpus' world that is used in the logic chapters, actually succeeds to stir an interest in the reader and then ....it is not followed up in the subsequent chapters such as the one on Bayesian networks)...
Some easy to grasp principles (such as basic propositional logic) are repeated ad nauseam while some difficult subjects (such as MCMC) are left as puzzling axioms, for us to decipher on our own.

I summarize my disappointment asking myself what I got from this effort that I have invested into this book, absorption and digestion wise, professionally speaking:
1. Did this book help me better understand the depth and breadth of the AI domain? - No.
2. Am I able to develop, even conceptually a plan for an AI application / "intelligent agent"? Absolutely not.
3. Did the book clarify for me the fields of logic, machine learning, reasoning, uncertainty, probability and so on? - No. I am as confused now as I was before embarking on this study project, maybe even more so.
4. Am I a smarter person, able to read now the multitude of scientific articles out there on the AI subject - after finishing this book? - No.

The only reason I gave it 2 stars instead of the single one it deserves - is because of the historical and bibliographical summaries the authors have nicely detailed at the end of each chapter. I've seen other books recommended in these reviews - and I intend to look into them shortly. CAVEAT EMPTOR (buyer beware) !

Form your own opinion on this book, don't let the gushing over this book force you into questioning your instincts

I thought I liked this book at first, but I had confused interest in AI with regard for this book.

Sure this was ground breaking. But, currently, it is bloated, full of wordy, unclear descriptions. I particularly dislike the coverage in: ch. 7, 8, 9 (logics + reasoning). ch. 13, 14 (prob, belief nets). Make the search chapters shorter, fewer. We get the idea, no need to spend so much time on it. Make the logic chapters shorter, dig deeper into those subjects if you want to use that much of the readers time. Scrap chapter 13 or write it over again (refer reader to Pearl's or others coverage of probability). It is partially to elementary, stating obvious rules with very simple usages. The rest of it jumps around, with unclear explanations. Chapter 14, skims past ideas, not enough time spent explaining ideas.

I particularly like the detailed references at the end of each chapter.

After glancing at Winston, Nilsson, and Poole books, I am leaning towards Poole, especially since I am more interested in the knowledge rep and reasoning than other areas.

Pros: Well organized, Description is clear and complete, good for beginners.
Cons: Examples chosen are not the best, author's attempts at humor are quite lame in most cases.