Traditional computer systems regularly fall short of the expectations of their designers. Holland, the inventor of the genetic algorithm, has devoted his career to the study of computational mechanisms that surprise us by giving back more than we put into them. In this book, he looks at the fundamental principles that enable systems to exhibit more complex and sophisticated behavior than their components do. The book is addressed to the general public, but mathematical sidebars give it sufficient rigor to be useful to researchers. It provides an elegant high-level view of the overall problem, often invisible at the level of individual technical projects, that promises to shape new research for years to come.

Two fundamental ideas form the backbone of Holland’s exposition of emergence, illustrated by board games and numbers. Board games highlight the notion of a system whose state changes over time, constrained by a set of relatively simple rules. The combinatoric potential of successive transitions means that even simple rules, as in go or chess, can yield a wide range of nonobvious behaviors. Numbers illustrate the idea of abstraction, in which most details are sheared away to leave a fundamental building block that is common to many situations. For instance, as different as two cars, two ideas, and two clouds are, they have in common two-ness. Research in emergence is driven by constructing models that abstract from real-world entities to their underlying building blocks, constraining the behavior of these blocks by rules, and then studying the overall system behavior when a computer executes the rules.

In moving from these two basic mechanisms to a general theory of emergence, Holland describes two experiments in some detail. The first is Samuel’s checkers program. Holland discusses the lessons it teaches about the nature of generated complexity, the importance of using predictions as a basis for learning in the absence of immediate reward, the need to give credit to actions that set the stage for later victories, and the need to anticipate the actions of other agents. The second experiment is Holland’s own early work in neural models of vision, revealing the power of persistent patterns of dynamic activity in a neural network.

Against this background, Holland develops the foundational construct of his theory, the constrained generating procedure, or cgp. A cgp is a recursively defined structure whose basic element is a transition function that maps an input vector and a current state to a new state. Individual cgps can be combined by mapping the state of one to an input of another, yielding an aggregate that is also a cgp. Although this basic cgp is computationally complete, more insight can be gained by using a representational extension that permits the interconnections between lower-level cgps to vary over time, yielding a cgp-v. Holland motivates the cgp formalism by using it to introduce cellular automata and their gliders (an example of emergence). Then he demonstrates the power of the cgp abstraction by using it to model Samuel’s checkers player, structures in the central nervous system, Mitchell’s CopyCat model of analogy formation, billiard ball physics, and genetic algorithms. The recursive nature of cgps focuses attention on the importance of different levels of description and the role of microlaws and macrolaws in understanding emergence.

The final chapters step back from this technical treatment to consider how emergence functions in the progress of scientific theory. Existing theories are building blocks that can be joined together to form new models of domains that are not yet fully understood. This process is analogous to the recursive linkages that generate layers of cgps. A critical question is how scientific innovators connect components from widely differing domains. Drawing from examples in both science and the arts, Holland suggests that the root dynamic of scientific innovation is closely connected to the mechanism of metaphor in natural language, which finds linkages between the auras that surround different concepts, thus enabling us to see each of the linked concepts in a new light. The book thus closes on itself recursively. It begins with the objective of developing a new scientific theory to explain emergence, and ends with a metatheory claiming that scientific theories themselves are instances of emergence.

Like many new scientific terms, “emergence” is widely used and abused in the popular press. Holland’s book brings the concept into focus in a way that will make it more accessible to the general public, and more precise and useful for researchers.