Artificial Intelligence: A Modern Approach, by Stuart Russell and Peter Norvig, is a university textbook now in its fourth U.S. edition (Pearson, 2021), and its companion site says it has been adopted by 1,556 schools in 134 countries or regions. The book is organized around the idea of an intelligent agent and runs from search and logic to machine learning, robotics, and a closing part on ethics and safety. It is written for an undergraduate course sequence, not for casual reading.
Key takeaways
- The fourth edition follows the 2010 edition and, by its preface, is about 25 percent new material.
- Machine learning gets more weight than hand-built knowledge engineering, and deep learning, probabilistic programming, and multiagent systems each have a chapter.
- Exercises moved online, with code implementations on GitHub, so the print book alone is incomplete as a course.
- The stated prerequisite is sophomore-level computer science, with some calculus and linear algebra helpful.
- The book is a survey with references, not a hands-on guide to building a current chatbot.
What the book is and who it is for
The preface describes the aim as conveying the ideas that have emerged over seventy years of AI research and two thousand years of related work. The authors say the book is primarily intended for an undergraduate course or sequence, that one semester can use selected chapters, and that a graduate course or self-study are also possible. The only prerequisite named is familiarity with algorithms, data structures, and complexity at a sophomore level. Readers expecting a quick introduction will find a large reference: the US edition index ends on page 1069.
The title's phrase "a modern approach" is explained in the preface. The authors chose to tell the story from a current perspective and to recast early work in today's terminology, and they apologize to researchers whose subfields became less recognizable as a result.
How the fourth edition is organized
The companion site lists seven parts, which makes the book easy to map onto a course plan.
| Part number | Title | Typical topics |
|---|---|---|
| 1 | Artificial Intelligence | Introduction, intelligent agents |
| 2 | Problem-solving | Search, complex environments, games, constraint satisfaction |
| 3 | Knowledge, reasoning, and planning | Logical agents, first-order logic, inference, automated planning |
| 4 | Uncertain knowledge and reasoning | Probability, probabilistic reasoning, decisions, multiagent decisions |
| 5 | Machine Learning | Learning from examples, probabilistic models, deep learning, reinforcement learning |
| 6 | Communicating, perceiving, and acting | Language processing, computer vision, robotics |
| 7 | Conclusions | Philosophy, ethics, and safety; the future of AI |
The preface of the Global Edition counts 29 chapters, each sized for about a week of lectures, so a full pass takes two semesters. The US site lists 28 chapters plus two appendices. The two counts differ, and the reason could not be confirmed from the sources.
What changed from the 2010 edition
The preface lists the changes itself. The authors shifted toward machine learning because of more data, computing resources, and new algorithms. They gave deep learning, probabilistic programming, and multiagent systems their own chapters, revised natural language, robotics, and vision in light of deep learning, and added robots that interact with humans. They also report that 22 percent of the citations are to works published after 2010.
The most conceptual change concerns the goal of an AI system. Earlier editions defined it as maximizing expected utility under an objective supplied by human designers. The fourth edition drops the assumption that the objective is fixed and known: the system may be uncertain about what the humans it serves actually want, and must still behave sensibly. That shift links the textbook to the safety and ethics material in the final part, and it appears in the preface as a deliberate departure.
Strengths, limits, and how to use it
The book's strengths follow from its design. It covers the field's classical core in one framework, supplies pseudocode, and ties topics together through the agent idea. Its limits are equally structural. A single print volume dated 2021 cannot track fast-moving areas, and the field has changed since, so the book is best treated as foundations to read alongside current papers and documentation. This is a description of the book's stated design, not a ranking against other textbooks, and the site hosts a separate reviews page for readers who want published opinions.
Practical notes for readers: the algorithm implementations are published in a code repository, and the exercises are on a website instead of at the end of chapters. The book's cover shows the final position of the 1997 Deep Blue match against Garry Kasparov and a Go position from the Lee Sedol match, a detail the preface explains.
FAQ
Is this book suitable for self-study?
The preface says it can be used for self-study or as a reference, though it is written for a university course. A reader should be comfortable with basic computer science and some mathematics.
Is the online material free?
The companion site hosts code, pseudocode, figures, and an exercises website. The book itself is a commercial publication from Pearson.
Does it cover large language models?
It covers deep learning and deep learning for natural language processing as chapters, based on the 2021 edition. Developments after that date are outside its scope.
How widely is it used in Canada?
The adoptions page counts 37 Canadian institutions, among them the University of British Columbia, McGill University, and the University of Toronto. The list depends on schools reporting their use.
Related reading
- A review of another introductory AI book
- Whether AI can work without data
- Pattern recognition in machine learning
- AI's role in modern society
More in AI Basics.
Last verified: October 2026
Sources
- Artificial Intelligence: A Modern Approach, 4th US edition, companion website, University of California, Berkeley, modified August 22, 2022
- Schools that have adopted AIMA, companion website, University of California, Berkeley
- Preface, Artificial Intelligence: A Modern Approach, 4th Global Edition, Russell and Norvig, Pearson, 2022
- AIMA exercises website, aimacode project