Series: SFI Studies in the Sciences of Complexity Pub. Date: July 2004 Publisher: Oxford University Press Cloth ISBN 0195165322 List Price $65.00 Paper ISBN 0195165330 List Price $35.00 Order Info More Info: Back of Book Info Table of Contents Introduction Series Info Return to book list

In a world of uncertainty, adversity, and rapid change, it might be thought that failure of systems, whether natural or designed, should prove the rule rather than the exception. And yet we are surrounded by examples of phenomena on all scales that we instinctively label as robust to failure, whether because of the agility with which they have responded to changing circumstances, or because of their resilience in the face of internal or external attack, or merely because they have proved so long-lived Robustness is a term that captures our intuitive sense of one of the key determinants of long-term successor failure, but what do we mean by robustness, and what specific features of a phenomenon contribute to its robustness or fragility?

In the past few years, the concept of robustness has been the subject of growing interest in the natural and engineering sciences. Building on traditional fields such as stability, reliability, and control theory, the study of robustness focuses on the ability of a system to maintain specified features when subject to assemblages of perturbations either internal or external (see, for example, Chapter 1 in this volume, and the web site ). In most cases, the system of interest is not in equilibrium, and the perturbations are typically such that it is unrealistic to attempt estimations of their supports and distributions.

Some questions of interest then include:

1. What is meant by Robustness in the various contexts in which the term is used? In what ways does robustness differ from stability, persistence, resilience, and recovery?

2. What are the origins of robustness? Do biological organisms evolve robustness? What is the "null hypothesis" regarding robustness; in other words, what does a functionally fit but non-robust system look like and how does it evolve?

3. What are the organizational principles--possibilities include spatial structure, redundancy, modularity, diversification, and hierarchy among others---that characterize highly robust entities? What are the costs of these organizational principles?

4. What are the consequences of robustness for evolvability adaptability, and degree of fitness of an entity to its environment?

Questions such as those listed above stimulate a rich set of research initiatives, and are contributing to new ways of thinking on issues ranging from the architecture of regulatory control, through the relation between performance and flexibility, to the evolution of general-purpose information-processing algorithms in these contexts. Examples of specific phenomena for which robustness is proving a useful "handle" include: the effects of structural mutation on cellular processes of metabolism and growth; directed evolution of enzymes with specified catalytic properties; disturbance regimes and recovery of ecosystems; computer network security systems subject to purposeful attack; and design of Internet protocolsystems. Insights that have been achieved in these contexts relate to:

1. The relationship between robustness and flexibility; in particular the role of robustness in reducing the potential lethality of mutations and increasing an organism's capacity to accumulate nonlethal genotypic and phenotypic variation. One consequence is to reduce the number of mutations needed to produce phenotypically novel traits [7, 9].

2. The robustness of food web networks that consist of a large number of weak interactions together with a small number of strong interactions. Endogenous and exogenous changes are then more likely to result in local elaboration of new morphologies rather than in a catastrophic failure of global organization [7].

3. Implications for robustness of what software engineers call "on-line management," namely, the need to maintain current functionality while implementing change or repair. Examples include the separation in protein evolution algorithms of modules governing function and structure so as to permit exploration of new functionality while maintaining necessary structure (see the chapter by Voigt et al.).

4. The existence of "congruence principles" that translate robustness at one level into robustness at another level. Examples include mutational robustness arising as a correlated response to selection on environmental robustness in RNA evolution [1], and the use of the same molecular mechanisms to realize both learning capabilities and developmental stability in neuronal systems [8].

5. The existence of conservation principles for robustness implying that systems that have evolved or been designed to withstand specified thresholds of shocks are commensurately susceptible to shocks above those thresholds and to other forms of shocks [3].

6. An analysis of biological sensory-processing systems that use redundancy of representation as a general-purpose algorithm and thereby gain multiple functionalities, including improved abilities to deal with poor signal-to-noise ratios, and to separate out multiple signal sources [11].

7. The identification of an extensive repertoire of biological mechanisms that employ diverse strategies that can be characterized as: (1) buffering the organism from the effects of an insult; (2) amplifying the insult to enable the organism to purge it; and (3) identifying and repairing the damage from the insult.

8. The establishment in ecosystem management of the need for policies and management that are flexible, adaptive, and experimental at scales compatible with the scales of critical ecosystem function [2, 6, 10], and recognition of the pathways by which fixed rules and rigid management structures lead to systems exhibiting a loss of robustness---i.e., that suddenly break down under disturbances that previously could be absorbed [5].

The purpose of the present volume is to present some of the recent advances made in the understanding of general principles of robustness especially in the context of evolutionary and developmental biology, ecology, and computer network design. Even more importantly, however, the volume attempts to provide examples of how researchers are trying to think about robustness. Each of the chapters is intended to highlight and to illuminate key issues in exploring robustness; namely, the tension between staying the same and responding to change; opportunities for innovation, and vulnerabilities to collapse, on multiple scales; effects of interactions among slow variables such as evolutionary change and fast variables such as ecological shifts; and, interwoven throughout, the role of adaptation and learning.

It is important to note, however, that while the volume attempts to survey and to integrate where possible diverse perspectives on robustness, it in no way seeks to unify all such perspectives, or to put forward "universality principles" for robustness that would be inconsistent with the patent diversity and distinctiveness of the range of processes to which the concept applies.

In fact, as the examples above indicate and the chapters that follow will illustrate in detail, diverse and sometimes conflicting interpretations of robustness are found to be useful at this point in different subcommunities in science and engineering. For example,

1. In engineering applications, robustness of systems is the object of intense effort, so much so that regulatory and control features designed for the explicit purpose of achieving robust performance often outweigh in both complexity and cost the actual functional features. Typically, robustness in these contexts is understood to mean reliability of function in the presence of failures with estimable probabilities and supports. Moreover, robust design is typically implemented "after the fact"; in other words, functionality comes first, with regulatory controls imposed subsequently to avoid failure modes. The incorporation of robustness as an integral component of the functional design process is a relatively new idea in engineering, and has yet to be realized to any real extent.

2. In the context of software engineering, robustness is distinguished from correctness. A program is expected to perform correctly on cases covered by its specifications (which might include faulty data, user error, etc.), but programmers also explicitly design for robustness outside of specifications---for example, the program might be expected to fail gracefully in the event of a disk crash. The problem of designing robust computer network systems (as discussed in the chapters by Forrest and by Doyle et al.) is an extreme example illustrating the challenge of protecting against what may be inherently unforeseeable.

3. Within developmental biology (see chapters by Lewontin and Goss, Krakauer and Plotkin, and Marder and Bucher), robustness typically refers to the ability of developmental processes to stay "on track" in the presence of perturbations such as environmental insultor developmental noise or knockout mutations. In recent years, robustness in cell biology has been used to describe the ability of certain metabolic and regulatory processes to perform correctly within a large range of parameters.

4. In the context of ecosystems robustness is often interpreted as what Gunderson and Holling (see the chapters by Webb and Levin, and by Walker et al. For more details) call "ecological resilience;" namely, the capacity of a system to undergo disturbance and still maintain its functions and controls. Note that no concepts of metrics or of equilibrium are implied in this definition of resilience, and the concept differs in this way from traditional notions of stability or other forms of ecosystem resilience, where the appropriate measure is the rate at which the system returns to equilibrium following a perturbation.

Essential differences notwithstanding, the twelve chapters in this volume can be said to adopt two complementary perspectives on robustness in biological, ecological, and engineering systems. The first (exemplified by ecological studies such as those by Webb and Levin, and Walker et al., and the neurophysiology work by Marder and Bucher) views robustness as characterizing a stage in the developmental history of a process. The process is analyzed as a set of dynamic interactions with feedback across multiple scales and in multiple dimensions on multiple networks. The question then is the role of these different dynamics in providing flexibility or rigidity in the response of the process to uncertainty and change, and in leading to any of the future possibilities of innovation, persistence, degradation, or collapse.

The second perspective (exemplified by the chapters by Krakauer, by Doyle et al., as well as by Marder and Bucher) treats the robustness of a process---whether it be cell growth, or Internet message routing---as a measure of its sensitivity to perturbations at a fixed point in time. To a large extent, this perspective treats robustness as a characteristic of a process that uses feedback control to perform distributed information-processing. The goal then is to understand the features such as error-correction or buffering that enable the process to perform successfully even with model uncertainty, unforeseeable consequences, conflicting data, and other complexities that could preclude the process from functioning as desired.

A challenge for future research in robustness is to use the insights being gained from perspectives such as those described above to generate actual design principles for robustness. A very different but equally important challenge is to construct new perspectives that incorporate useful aspects of the "developmental" and the "feedback control" views described above, but that are relevant to social processes. Any such perspective would necessarily include the uniquely ssocial features---including the critical role of cognition and learning, intentionality and identity, evolving cultural repertoires, and the extraordinary human capacities for effective behavior including deliberate collective action or the envisioning of alternative realities---that lend social processes their distinctive flavor of complexity, and that are clearly key to any study of social robustness.

Bibliography

1

Ancel, L. W., and W. Fontana. "Plasticity, Evolvability, and Modularity in RNA." J. Exper. Zool. 288 (2000): 242--283.

2

Costanza, Robert. Institution, Ecosystems, and Sustainability. Boca Raton, FL: Lewis Publishers, 2001.

3

Doyle, J., and J. Carlson. "Highly Optimized Tolerance: Robustness and Power Laws in Complex Systems." Phys. Rev. E. 60 (1999): 1412.

4

Gumerman, G., and T. Kohler. "Creating Alternative Cultural Histories in the Prehistoric Southwest." In Proc. Durango Conference on Southwestern Archaeology, 1995.

5

Gunderson, L. H., and C. S. Holling. Panarchy: Understanding Transformations in Human and Natural Systems. Washington, DC: Island Press, 2001.

6

Gunderson, L. H., C. S. Holling, and S. S. Light. Barriers and Bridges to the Renewal of Ecosystems and Institutions. New York: Columbia University Press, 1995.

7

Kirschner, M., and J. Gerhart. "Evolvability." PNAS 95 (1998): 8420--8427.

8

Marder, E., and R. L. Calabrese. "Principles of Rhythmic Motor Pattern Generation." Physiol. Rev. 76 (1996): 687--717.

9

van-Nimwegen, E., J. P. Crutchfield, and M. Huynen. "Neutral Evolution of Mutational Robustness." PNAS 96 (1999): 9716--9720.

10

Walters, C. J. Adaptive Management of Renewable Resources. New York: Macmillan, 1986.

11

Zweig, G. Private communication.