John Holland is one of the inventors of computer algorithms that mimic genetic evolution. He delivered the first lecture of the Santa Fe Institute’s annual series in honor of Stanislaw Ulam; this book is the archival record of those lectures.
A recurring theme in complex adaptive systems (CAS) research is that interactions of small, relatively simple entities may yield much more complex behavior on a larger scale. In that spirit, Holland seeks to develop low-level models that may be useful for understanding large-scale issues, such as the dynamics of a city or of a viral epidemic. The first and fifth chapters outline this general program, while the second, third, and fourth summarize Holland’s own work in support of this agenda.
Chapter 1, “Basic Elements,” introduces the challenge of understanding the behavior of large, complex systems. It observes that such systems are invariably composed of large numbers of active elements or “agents,” and then synthesizes seven basic characteristics of these agents and the systems they compose:
Agents aggregate their behavior and thus form higher-level meta-agents.
This aggregation is guided by tagging, the ability to respond differentially to other agents based on some sort of marker.
The behavior of agents and their aggregates is nonlinear, leading to complexity and unpredictability.
Their interactions are characterized by flows, which multiply the impact of individual agents and provide recycling and feedback.
Agents are diverse from one another, thus mutually defining behavioral niches.
They contain internal models (in the simplest cases, coded in their genes) that enable them to predict and respond appropriately to the environment.
These models are made up of reusable building blocks.
Chapter 2, “Adaptive Agents,” is a detailed but technically simple explanation of the basic ideas of Holland’s classifier systems and genetic algorithms, two contributions that established his reputation and have inspired much subsequent work in artificial life. These ideas provide an example of a simple model for the individual agent in a complex adaptive system. Starting with the notion of a rule-based system, the chapter develops two mechanisms--credit assignment and genetic rule discovery--by which an agent can adapt to its environment, and leads even a nontechnical reader to the schema theorem, which is the fundamental theorem of genetic algorithms.
Chapter 3, “Echoing Emergence,” turns from individual agents to a framework for modeling a system of such agents. This chapter and the next describe Echo, an embryonic research program that explores six successively more complex models of agent interaction. Agents can recognize one another as friend or foe, based on an evolvable chromosome that includes externally visible tags. Agents exist in an environment that supplies resources and defines what it means for agents to be close enough to one another to interact. Agents reproduce when they have collected enough of the appropriate resources to duplicate their chromosomes. Extensions to this basic model include mechanisms to impose conditions on the interaction between two agents, to transform resources from one type to another, to permit individual agents to aggregate with one another, to impose conditions on the selection of mates, and to control whether an agent replicates. These mechanisms were inspired by embryogenesis, in which an initial single cell divides and specializes to form a complex multicelled organism with specialized components.
Chapter 4, “Simulating Echo,” illustrates how to simulate a collection of Echo agents by applying Echo to the Prisoners’ Dilemma. Holland summarizes two broad uses of the simulation of Echo communities: as a basis for thought experiments to explore the principles of complex adaptive systems, and to permit people to experiment with naturally occurring complex adaptive systems and thus learn to manage them more effectively.
Chapter 5, “Toward Theory,” articulates the need to move beyond simulation to a mathematical theory if one is to achieve useful understanding and reliable insights. Holland’s initial sketch of such a theory links a lower tier, in which agents mingle freely and rapidly and are governed by flows, to an upper tier governed by much slower time scales, in which adaptation relies on evolution. Holland suggests that extending this foundation will require interdisciplinary effort; computer-based thought experiments; a correspondence principle that aligns the new theory with the results of prior studies in relevant disciplines; and a mathematics of competitive processes based on recombination.
As befits a popular lecture series, this book is easily accessible to the educated layperson, and the short bibliography distinguishes books understandable to the general reader from more technical works. In spite of Holland’s popular style, this volume is a technically reliable introduction to some of the leading concepts and approaches in the new field of complex adaptive systems, and will be a useful reference for scientists and laypeople alike who are interested in Holland’s work.
