In a clear yet formal way, this book illustrates how intelligent agents are conceived from the different perspectives that dominate the artificial intelligence (AI) field today, and according to the nature of the task to be performed. This basic goal of the book, present since the first edition, makes it particularly useful as an introductory textbook and a cornerstone of the study of AI in undergraduate and graduate courses.
A unified vision of the fields is also presented in the book, with explanations of the theory and basic methods of the different approaches that are dominating AI now and look most promising. The organization of the book is consistent with the objective of being used as a textbook. Each chapter presents the concepts; illustrates the methods derived from the theory; comments on the available references, including other books, software, and Web resources; and poses programming exercises. The book is complemented by a Web site (http://aima.cs.berkeley.edu) with resources for instructors, reviews, source code, and a very active forum.
I used the first edition as the textbook for several AI courses, finding it flexible enough to be used in courses with different lengths, emphasis, and orientations. I am glad to see all the errors that the community found in the first edition corrected; when I received the book, I ran to the neural network learning section to check a comment that I made on the backpropagation algorithm presented in the first edition. I found not only the algorithm corrected, but improved organization, better explanations, and new exercises.
The book covers areas that are sometimes ignored in other books, such as reasoning under uncertainty, learning, natural language, vision, and robotics, and explains in detail some of the more recent ideas in the field, such as simulated annealing, memory-bounded search, global ontologies, dynamic belief networks, neural nets, inductive logic programming, computational learning theory, and reinforcement learning.
In this second edition, the coverage on probabilistic learning is extended, including Bayesian networks. Markov decision problems are also more comprehensively presented, solving the deficiencies presented in the first edition. New algorithms are introduced in all the sections, such as graphplan in planning, boosting and kernel methods in learning, and so on.
A new section on the ethical issues and risks of AI is proudly presented by the authors in the chapter devoted to philosophical foundations, but I think that including this material using the same format of the rest of the book, including exercises, is not appropriate. This subject obviously needs to be discussed, but from a higher perspective that does not fit in the format of the other chapters.
The book is well written, the typography is excellent, and the illustrations are clear and truly contribute to complete understanding of the material. The bibliography was revised, improving the already outstanding one presented in first edition.
I strongly recommend the book as a textbook for introductory to intermediate level courses in AI, and also as a reference to the main methods used in AI today.