Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Artificial intelligence: foundations of computational agents (2nd ed.)
Poole D., Mackworth A., Cambridge University Press, New York, NY, 2017. 760 pp. Type: Book (978-1-107195-39-4)
Date Reviewed: Jan 23 2018

Artificial intelligence (AI) can be defined as the study of the design of intelligent computational agents. Unfortunately, this is a recursive definition caused by the lack of a consensus on what intelligence really means. In the past, it could refer to being able to do simple mechanical tasks such as arithmetic, showing some proficiency in more complex domains such as chess, or identifying objects within images, a task where computers have surpassed human-level accuracy only recently with the help of deep learning techniques. Since we are dealing with a moving target, I always preferred Rich and Knight’s definition of AI as the study of how to make computers make things that, at the moment, people do better [1].

Poole and Mackworth, from the University of British Columbia in Canada, try to be more systematic and provide a general framework that can be used to put a myriad of AI techniques in context. Their framework is organized around ten dimensions that define the design space of intelligence agents. It is suitable for providing a roadmap for the study of the infinitude of algorithms, heuristics, and approaches that comprise the AI landscape (a countably infinite set of them, to be rigorous). Their current ten dimensions expand the nine dimensions they used in their previous edition: modularity, representation scheme, planning horizon, sensing uncertainty, effect uncertainty, preference, learning, number of agents, and computational limits [2]. To those nine dimensions, they have added “interaction” in order to distinguish agents that reason online (while interacting with their environment) from agents that reason offline (before interacting with their environment).

Once the formal stage is set, Poole and Mackworth provide a general overview of the broad AI field from an interesting pedagogical perspective. Rather than devoting independent chapters to search algorithms, logical reasoning, planning, probabilistic models, machine learning, and so on, as common AI textbooks typically do, they introduce those techniques incrementally, from the simplest to the more complex.

First, they describe reasoning, planning, and learning with certainty (namely, heuristic search, constraint-based search, traditional planning, and supervised machine learning).

Later, they introduce uncertainty in the design of computational agents, which justifies the use of probabilistic models for reasoning (belief networks and Markov models), as well as Markov decision processes for planning (for example, value and policy iteration). Additional machine learning techniques are also included in this part of the book, including shallow overviews of unsupervised learning, probabilistic learning, and Bayesian learning. Prospect theory (an economic theory for modeling human preferences that matches empirical observations better than a purely rational utility-based perspective), game theory, group decision making (including Arrow’s impossibility theorem), and mechanism design are some of the less common topics in AI textbooks that are also covered by the authors. Reinforcement learning, a technique based on Markov decision processes that helps computational agents learn how to act, also gets the attention it deserves in its separate chapter, which describes Q-learning, SARSA, and SARSA with linear function approximation (the same approach that is employed in deep reinforcement learning, where the linear function approximation is replaced by an artificial neural network).

Finally, the last part of the book before the authors’ retrospective deals with reasoning, learning, and acting with individuals and relations. Readers might be surprised that first-order logic does not appear earlier in this book. Here, a peculiar approach to natural language processing is discussed based on definite clause grammars. A strange assortment of ontologies, meta-interpreters for the implementation of knowledge-based systems, inductive logic programming, and collaborative filtering round up the extensive survey of AI techniques this book provides.

A final chapter mentions the social and ethical consequences of AI-based systems. Even though I cannot agree with every argument the authors provide, nor many of the proposed solutions that range from additional legal regulations to redistributive wealth taxes and a universal basic income, it is useful that undergraduate students explore and discuss the nontechnical ramifications of the technical feats they can achieve in their work as engineers.

This textbook is accompanied by a website, http://artint.info/, where you can find its full text, software tools, and additional learning resources. Its most noteworthy feature is the peculiar order in which concepts are introduced for pedagogical purposes. This unique ordering accentuates the appearance of occasional forward references, which I always find annoying. On the other hand, it also paves the way for establishing many insightful connections between apparently disparate topics. Many interspersed examples and slide-like bullet points facilitate understanding, at the cost of some jumps and continuity breaks that hinder its legibility. Algorithms are provided in Pascal-like pseudocode, with some peculiarities such as nondeterministic constructs. Each chapter is framed by an intelligent opening quotation, a curated selection of references for further reading, and some proposed exercises for reinforcing the many concepts discussed in this textbook. Some of them are thoroughly covered from scratch, so that anyone with a basic background in computer science can understand them; many are just cursorily overviewed, so that students can situate them in their proper context, even though they might require additional references if they want to put them into practice. In summary, this textbook provides a decent alternative to Russell and Norvig’s hefty and almost ubiquitous AI textbook [3], with a strong focus on symbolic approaches and easy-to-understand descriptions of many probabilistic techniques.

More reviews about this item: Amazon

Reviewer:  Fernando Berzal Review #: CR145800 (1804-0171)
1) Rich, E.; Knight, K. Artificial intelligence (2nd ed.). McGraw-Hill, New York, NY, 1991.
2) Poole, D. L.; Mackworth, A. K. Artificial intelligence: foundations of computational agents. Cambridge University Press, New York, NY, 2010.
3) Russell, S.; Norvig, P. Artificial intelligence: a modern approach (3rd ed.). Pearson, Edinburgh Gate, UK, 2009.
Bookmark and Share
  Reviewer Selected
Featured Reviewer
 
 
Multiagent Systems (I.2.11 ... )
 
 
Applications And Expert Systems (I.2.1 )
 
 
Learning (I.2.6 )
 
Would you recommend this review?
yes
no
Other reviews under "Multiagent Systems": Date
Engineering intelligent hybrid multi-agent systems
Khosla R., Dillon T. (ed), Kluwer Academic Publishers, Norwell, MA, 1998. Type: Book (9780792399827)
Aug 1 1998
Linguistic geometry: from search to construction
Stilman B., Kluwer Academic Publishers, Norwell, MA, 2000.  395, Type: Book (9780792377382)
Jan 1 2001
 Transactional agents: towards a robust multi-agent system
Nagi K., Springer-Verlag New York, Inc., New York, NY, 2002.  205, Type: Book (9783540430469)
Apr 13 2004
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy