Would you like to create an intelligent agent that can interact autonomously with its environment and learn to improve its interactions? If so, how would you go about it? The answer lies in the field of artificial intelligence (AI). About 30 years ago, AI was a growing area, with predictions of broad capabilities and human-level intelligence not too far off. Then, the reality of its complexity and the limitations of available processing power sank in, and experts began to practice in deep rather than broad areas. To be sure, while there have been some notable successes in such areas--including medical diagnosis, expert systems, military robots, and even video games--the general-purpose system remains decades away.

This book is a wonderful, well-written introduction to a tremendously complex field that is interesting to many and fascinating to some. The authors manage the complexity by beginning with the simplest elements and building on these to progressively broaden and deepen the treatment. They provide a large number of references for those who wish to go beyond the text.

The concept of “agent” is a good choice: it can be easily understood and applied to many areas, and it embodies all of the elements of what a general AI system might need. That said, readers should have some familiarity with theoretical computer science and mathematical concepts, especially discrete mathematics, functions, factors, and arrays, as well as relations and relational algebra. Although the appendix provides definitions for these terms, definitions are no substitute for a more formal exposure to the ideas.

The book begins with some background on AI, what an agent is, representations of knowledge, hierarchical control, and acting with reasoning. The next part, “Representing and Reasoning,” covers what an agent with full knowledge of its environment can do, including chapters on states and searching, features and constraints, propositions and inference, and reasoning under uncertainty. This last chapter includes a development of probability theory and a treatment of beliefs in terms of probability.

The next logical step is that, while an agent begins with incomplete knowledge, it has the ability to learn from observation and optimize its own performance. In this part, the authors also cover learning and planning, including models of learning; representing states, actions, and goals; planning under both certainty and uncertainty; multiagent systems and game theory; learning belief networks; and reinforcement learning.

From this point on, the text considers reasoning about individuals--using a broad definition of “individual”--and relations. It includes a discussion of symbols and semantics, the relational rule language Datalog, proofs, applications in natural language processing (NLP), ontologies and knowledge-based systems, and relational planning, learning, and probabilistic reasoning. The book concludes by revisiting the dimensions of complexity and discussing social and ethical consequences.

As I noted before, the book is very clearly written. It presents the topics in a highly pertinent order, which allows readers to build upon previously learned concepts. Readers will appreciate the example AI systems that are carried forward through the text; as new and more difficult concepts are discussed, the examples apply to a system that the readers are already familiar with. The only nitpick I have is the text’s frequent references to the book’s Web site (http://www.cs.ubc.ca/~poole/aibook/); although it’s not specified in the text, I was able to find it. Furthermore, a related Web site (http://aispace.org/) provides implementations of some of the AI concepts discussed in this text.

As the authors state at the beginning,

This book provides the first accessible synthesis of the field aimed at undergraduate and graduate students. ... [It] balances theory and experiment, showing how to link them intimately together ... [and it] develop[s] the science of AI together with its engineering applications.

I agree wholeheartedly with these claims and recommend this excellent work.