The Master Algorithm

July 1, 2015
Traditionally, the only way to make a computer execute a task is to write precise instructions: an algorithm. As the author notes in this enthusiastic but not dumbed-down introduction to machine learning, it is impossible to "write a program to tell a computer how to drive a car or decipher handwriting, but if we give enough examples to a computer running a learning algorithm, it will figure out how to do it on its own." The ultimate learning program, writes Domingos (Computer Science/Univ. of Washington), is the master algorithm, and the process is well underway to allow computers to function creatively. Data alone is not enough. Defeating the world's greatest chess or Jeopardy players was a matter of brute force, but simpler computers running learning programs already beat talent scouts in baseball, connoisseurs in wine tasting, and doctors in disease diagnosis. Though lucid and consistently informative, Domingos' explanation of how a variety of scientific schools approaches the master algorithm requires close attention from readers. Symbolists believe that intelligence emerges from manipulating symbols, just as mathematicians solve equations by replacing expressions with other expressions. Connectionists try to reverse-engineer the brain. Evolutionaries write programs that change in ways similar to natural selection. Bayesians know that all learned knowledge is uncertain, so they emphasize 18th-century English clergyman Thomas Bayes' theorem, which can handle probabilistic inference. Finally, analogizers search for similarities in data and write code that combines them to make new predictions. "Armed with your new understanding of machine learning," writes the author, "you're in a much better position to think about issues like privacy and data sharing, the future of work, robot warfare, and the promise and peril of AI." With wit, vision, and scholarship, Domingos describes how these scientists are creating programs that allow a computer to teach itself. Readers unfamiliar with logic and computer theory will have a difficult time, but those who persist will discover fascinating insights.

COPYRIGHT(2015) Kirkus Reviews, ALL RIGHTS RESERVED.

Starred review from August 1, 2015
In the amazing machine logic that helps Netflix users find movies, eHarmony clients find dates, and Amazon customers find books, Domingos sees merely the forerunner of a powerful Master Algorithm that may soon revolutionize our world, making computers and robots self-programming learners that no longer need human guidance. Himself a pioneer in the field, Domingos tutors nonspecialist readers in the workings of algorithms that render a computer capable of teaching itself how to perform a task with ever-increasing reliability. Without delving into the complexities, readers explore the thinking of the five different groups striving to perfect the three fundamental elementsrepresentation, evaluation, and optimizationdriving this process and consider ways that creative software engineers might fuse the strengths of these five groups' divergent perspectives into one all-potent universal algorithm. When finally forged, that algorithm, Domingos predicts, will so thoroughly transform the world that human soldiers will disappear from battlefields, children will grow up with robot nannies, and ordinary Janes and Joes will use computer-assisted DNA design to claim the movie-star bodies they crave. Skeptics may dismiss Domingos' vision as hopelessly utopian, and libertarians may brand it as dangerously intrusive. But no one will find it boring. An exhilarating venture into groundbreaking computer science.(Reprinted with permission of Booklist, copyright 2015, American Library Association.)