Upload
euel-elliott
View
216
Download
0
Embed Size (px)
Citation preview
Chaos, Solitons and Fractals 20 (2004) 63–68
www.elsevier.com/locate/chaos
A complex systems approach for developing publicpolicy toward terrorism: an agent-based approach
Euel Elliott *, L. Douglas Kiel
School of Social Sciences, The University of Texas at Dallas, P.O. Box 830688, Richardson, TX 75083, USA
Abstract
This paper examines terrorist ‘‘fluids’’ as complex adaptive systems. The principles of agent-based modeling are
applied to global terrorism as a basis for developing agent-based models of terrorist behavior. Recommendations for
agent-based models of terrorist behavior as policy analytic tools are presented.
� 2003 Elsevier Ltd. All rights reserved.
Perhaps no greater challenge faces policy makers and policy analysts than how to design strategies and tactics to
protect against global terrorism. Protecting against the recurrence of terrorist attacks, such as those that befell the
United States in September of 2001, calls for developing an expanded appreciation of the incredibly complex and global
environment in which public policy decision making occurs. Clearly, the sciences of complexity afford the intellectual
lenses and methodologies for handling a task of such scope and scale. The study of complex adaptive systems (CAS)
allows us to gain a better understanding of both micro- and macrolevel social and political dynamics that may shape
and influence terrorist opportunity structures.
This paper explores ways in which both the concept of CAS and methods for studying CAS can be successfully
exploited to develop a better understanding of terrorist dynamics. Terrorist organizations are viewed as CAS for the
purposes of this analysis. We then discuss the use of agent-based simulations as a means for modeling the various
dynamics that must be understood in order to develop successful strategies of response to terrorism. We conclude with
some observations about the potential of agent-based simulation as a policy tool for combating terrorism and with
some observations concerning the challenges of complexity raised by global terrorism.
1. Terrorism as a complex adaptive system
Students of the sciences of complexity will readily recognize the applicability of the concept of CAS to terrorist
organizations, especially in the case of groups as such as Al-Qaeda. Consider the following definition of CAS by Dooley
[1]:
A CAS behaves/evolves according to three key principles: order is emergent as opposed to pre-determined, the sys-
tem’s history is irreversible, and the systems’ future is often unpredictable. The basic building blocks of the CAS are
agents. Agents are semi-autonomous units that seek to maximize some measure of good, or fitness, by evolving over
time. Agents scan their environment and develop schema representing interpretive and action rules. These schema
are rationally bounded: they are potentially indeterminate because of incomplete and/or biased information; they
* Corresponding author.
E-mail address: [email protected] (E. Elliott).
0960-0779/$ - see front matter � 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S0960-0779(03)00428-4
64 E. Elliott, L.D. Kiel / Chaos, Solitons and Fractals 20 (2004) 63–68
are observer dependent because it is often difficult to separate a phenomenon from its context, thereby identifying
contingencies; and they can be contradictory. Schema exist in multitudes and compete for survival.
This definition of CAS speaks directly to much of what is known about terrorism such as that evidenced by Al-
Qaeda. Terrorists as CAS can be seen as ‘‘semi-autonomous’’ agents. Terrorists often possess ‘‘schema’’ for action rules
with unpredictable results for even the terrorist is unsure of the outcomes of the intended event. The notion that schema
are also observer and context dependent reinforces the notion that what defines ‘‘terrorism’’ is often in the eye of the
beholder. And clearly, the ‘‘war on terrorism’’ concerns a deep battle over the dominant and ‘‘surviving’’ schema.
The current and established metaphor for understanding terrorist organizations is the network model. Thus analysts
focus on what is known about ‘‘networks’’ as a means for developing methods for combating these so-called networks.
While networks do indeed represent an important area in complexity analysis, another metaphor from the complexity
sciences may be more appropriate when considering the nature of terrorist organization. Studies from fluid mechanics
may provide a better understanding of global terrorism than the network model. Consider the value of replacing the
network model with the ‘‘fluids’’ model. Understanding terrorism as a fluid may provide improved insight into the
challenges of global terrorism and to the policy demands necessary for combating this global fluid.
Mol and Law [2] offer the concept of fluid as a means for distinguishing between spatial patterns within global
systems. Mol and Law see networks as crossing diverse regions where relations between the elements in the network
remain constant. Mol and Law’s concept of fluids, however, provides a more powerful vision of the nature of global
terrorist action and movement. Mol and Law define fluids as, ‘‘. . .neither boundaries nor relations mark the differencebetween one place and another. Instead, sometimes boundaries come and go, allow a leakage or disappear altogether,
while relations transform themselves without fracture. Sometimes then social spaces behave like a fluid’’ [3]. Thinking of
terrorist organizations as a fluid helps us to further understand the nature of the challenge that exists. While networks
may be depleted by the destruction of dominant nodes, fluids may simply move like jello under the weight of efforts to
crush them.
Understanding the nature of the policy challenge in combating the global fluid of terrorism requires a review of the
organizational structures used to combat terrorism. Naturally the dominant structure is industrial age bureaucracy.
Table 1 below contrasts the system components and system demands of bureaucracy versus the elements of a CAS, in
this case, the global terroristic fluid. While the contrasts between bureaucracy and CAS in Table 1 are perhaps overly
stark, an important point is made. There is a fundamental mis-match between the organizational structure aimed at
combating terror and the nature of the global fluid of terror. Consider that the institutional response of the United
States to the attacks of September 11, 2001 was to generate the largest re-organization in US governmental history
producing a massive bureaucracy with an almost unfathomable organization chart. One could imagine some kind of a
density and mass measure of such bureaucracy perhaps labeled ‘‘Weberian weight’’. Throwing massive Weberian
weight at a phenomenon with the ‘‘squishiness’’ of jello may simply push the jello into unpredictable directions––yet the
jello survives.
Table 1 also emphasizes the dynamic elements of CAS relative to the overarching stability function of bureaucracy.
We thus begin to see the fundamental nature of the conflict between the objects of terror and the producers of terror.
While terror at its core hopes to invoke instability and uncertainty, the mechanisms for contending with terror aim to
impose stability and certainty.
This fundamental clash between mechanisms that seek instability and those that seek stability must serve as an
intellectual foundation for studies of terrorism and for public policy solutions. From the perspective of the West,
Table 1
System components/demands of bureaucracy and terrorist fluids as CAS
System components/demands Bureaucracy Terrorist fluids as CAS
Individual behavior Rule-bound Adaptive rules
Organizational structure Hierarchical Fluid mechanics
Integration of system components Tightly coupled Loosely coupled
Systems maintenance Equilibrium dynamics Dis-Equilibrium dynamics
Desired state Stability Instability
Hierarchical dynamics Linear Nonlinear
Evolutionary time scale Incremental adaptation Punctuated adaptation
Evolutionary fitness regime Stability The Edge of chaos
Decision criteria assumptions Certainty Uncertainty
Future orientation Predictability Unpredictable
E. Elliott, L.D. Kiel / Chaos, Solitons and Fractals 20 (2004) 63–68 65
crushing instability may lead to threats to the flexibility mechanisms of civil rights and freedoms that distinguish these
cultures from their historical antecedents and contemporary competitors.
2. New approaches to simulation
Recent advances in simulation methods, in concert with novel views of the emergence of social phenomena, are
providing new methods for modeling terrorist behavior. Improvements in these modeling approaches have generated a
growing interest in the simulation of a variety of social phenomena [4]. One of the emerging research approaches within
the field of social simulation is agent-based modeling (ABM). Agent-based or computational models represent efforts to
create computer-based ‘‘microworlds’’ in which individual agents, with diverse qualities, interact on some defined
environment labeled a landscape. These microworlds or ‘‘would-be-worlds’’ [5] become the foundation for producing
‘‘artificial life’’ in the computer. In the case relevant to this paper, a variety of human social environments and settings
where terrorism may express itself, can be simulated.
ABM range across fields as diverse as labor economics [6] to criminology [7]. Management scholars are beginning to
use ABM to simulate behavior in organizations [8]. And, political scientists [9] have used ABM to examine the creation
and destruction of international alliances. Potential applications of ABM range across all human social settings in
which simulated agents interact with one another in some defined space. Such approaches to explaining the dynamics of
CAS have increasingly captured the attention of the scientific community [10].
3. Principles of agent-based models
ABM abides by a novel view of the creation of emergent human social behavior. It is premised on the notion that
emergent social phenomena are created by the interactions of various microlevel agents on landscapes. This perspective
is an alternative to the conventional social science view that the hierarchies of social structures are primary to social
interaction. This perspective led Epstein and Axtell [11] to label ABM as ‘‘social science from the bottom up’’ as
compared with the ‘‘top-down’’ view of traditional social science. Epstein and Axtell see ABM as ‘‘generative social
science’’ because a goal of this perspective is to examine how social phenomena emerge ‘‘from the bottom up’’ via the
interactions of multiple agents on a landscape. The logic of this approach has obvious implications for reconsidering
social phenomena such as terrorism in the case of Al-Qaeda. Is Al-Qaeda-based terrorism the result of some imposed
structure, or is the structure of this form of terror generated from the bottom-up via the interactions of multiple in-
dividual agents?
Several basic principles guide the development of agent-based models. The earliest and still relevant presentation of
these principles was first offered by Langton [12]. These principles, when viewed as behaviors, have much to tell us
about the potential actions of terrorist agents and the applications of ABM to the study of terror.
1. The model consists of a population of simple agents.
2. There is no single agent that directs all of the other agents.
3. Each agent details the way in which a simple entity reacts to local situations in its environment, including encounters
with other agents.
4. There is no rule in the system that dictates global behaviors.
5. Any behavior at levels higher than the individual agents is therefore emergent.
Langton’s first principle emanates from the general assumption within the sciences of complexity that ‘‘simple rules
can generate complex behaviors and structures’’. Agents, such as individual terrorists may not require a comprehensive
set of rules to guide their behavior. Rather, a few simple rules are all that is necessary. Examples of such simple rules
might be ‘‘always create the maximum mayhem given available resources’’ or ‘‘leave the scene if you believe you are
being followed’’. ABM efforts have already shown the capacity for simple rules to generate novel forms of structure and
organizations [13]. In Epstein and Axtell’s Sugarscape model, for example, ‘‘. . .each individual agent has simple rulesgoverning movement, sexual reproduction, trading behavior, combat, interaction with the environment, and the
transmission of cultural attributes and diseases’’ [14].
Langton’s second principle reveals the lack of pre-existing social structure embedded in agent-based models. This
principle also allows for the creation of ‘‘autonomous agents’’ free to employ their strategies that are enacted by simple
rules. In short, there is no assumed hierarchy or systems of dominance. This is particularly relevant to terrorists who are
‘‘embedded’’ and who may need little, if any, hierarchical support.
66 E. Elliott, L.D. Kiel / Chaos, Solitons and Fractals 20 (2004) 63–68
The concept of ‘‘local situation’’ is essential to understand Langton’s third principle. Autonomous agents are acting
under their own rule-driven volition and respond only to those stimuli in their immediate environment. In short, agents
cannot see but just a few pixels ahead of themselves in the virtual computer-based world. This understanding of the
limited vision of agents can be seen as a means to interject the notion of bounded rationality into these simulations.
Thinking strategically is difficult. Agents are forced to confront immediate challenges and the results are always subject
to risk and uncertainty. But in the case of terrorists this leaves cells free to adjust their tactics.
Langton’s fourth principle can be redefined to note that no overarching (global) rule determines agent behavior.
From this perspective, agent behavior is local as each agent senses and responds to local stimuli. This principle is
essential to understanding that generative social science assumes that structure is created by the interactions of agents.
There is no central guidance system with ABM. If a central guidance system were to be generated it would be the result
of the interactions of agents. This insight is consistent with the recognition of the element of surprise in terrorism. The
objects of terrorism rarely know what to expect and are always kept guessing.
This final principle is perhaps the most important of Langton’s principles. Behavior at levels higher than the in-
dividual agent reveals that some form of structure has ‘‘emerged’’ that was not predefined by the rules of the various
agents. The concept of ‘‘emergence’’ is central to understanding ABM. Through their interactions, agents may generate
‘‘social macrostructures’’ that emerge not as a result of a global plan but because of the actions of multiple agents. In
fact, one can then imagine that an agent-based model with terrorists and counter-terrorists may create a ‘‘global
macrostructure’’ in which counter-terrorists converge in a manner consistent with the current international effort to
fight terror.
4. Visual and data representation of agent-based models
ABM is both a visual and a data representation medium. One of the essential elements of ABM is the ability to alter
the parameters of the model to test various outcomes and potentialities of the model. Once the simple rules for agents
are defined and the landscape for agent action is determined, the parameter settings serve to delineate elements of the
model such as the availability of landscape resources. Changing the parameters such as the upper and lower bounds of
an agent’s capacity to engage in a variety of behaviors potentially alters model outcomes.
ABM models are thus dynamic as each iteration produces various levels of agents of different types. ABM models
also differ from traditional models by focusing on continuous dynamics over time. Traditional modeling generally seeks
to find some equilibrium representing the ‘‘solution’’ to the simulation. ABM seeks to interject a stronger element of
time by allowing the simulation to continue over a large range of iterations. As long as the dynamics continue in the
ABM, there is no reason to shut the model down. There may simply be no single or optimal solution to an ABM
simulation. The goal is to examine dynamics over time without the expectation that the simulation will ever ‘‘settle
down’’. What may seem like an equilibrium, or an overwhelming victory, for certain agent types may dissolve as that
agent type overreaches its ‘‘fitness’’ on the landscape. Emergence thus is not necessarily a singular or stable phe-
nomenon. Structure may emerge over various times during a run of the model thus generating a larger concern for
patterns of change. If the world is a continuously running dynamic model, then should our simulations not attempt to
emulate this evolving dynamism?
5. Agent-based modeling as a policy tool for combating terrorism
At the macrosocial level, a wealth of issues emerge. These include questions of strategies for ferreting out potential
terrorists and terrorist cells. What approaches can be used to discourage cells from successfully forming, or if formed,
carrying out terrorist operations? At the same time, policies need to be developed that do not do harm to our traditions
of civil rights and at the agency and organizational level, we require new modalities of bureaucratic design and behavior
to cope with the realities of terrorism. The rigid bureaucratic structure, the existence of which has become all too
obvious, must draw upon the lessons of versatility and adaptability; they must, in some sense, serve to mirror the
adaptability and versatility of the very terrorist organizations they seek to eliminate.
The agency or organizational level of analysis is crucial because it serves as a linchpin between the macro- and
microlevels of operations. It is the responsibility of the organizations charged with developing and carrying out policy
to manage both the systemic levels of concern as well as the microlevel field behavior of both terrorists and anti-terrorist
counterparts, and to understand the interaction between the levels of analysis and concern. This understanding leads to
several basic questions that agent-based simulation of terrorism must ask.
E. Elliott, L.D. Kiel / Chaos, Solitons and Fractals 20 (2004) 63–68 67
1. Which types of terrorists (rules) agents grow and expand under varying landscape conditions?
2. Which types of terrorists (rules) shrink and decline under varying landscape conditions?
3. How do either static or dynamic terrorist rules alter outcomes?
4. What kinds of macrostructures emerge contingent on the mix of bureaucratic responses to terror?
5. Is there a stable winning strategy against terror under all landscape conditions?
These types of simulation may thus provide an improved means for assessing the relative risk and uncertainty
generated by various anti-terror strategies. Consider the risk and uncertainty involved in any large-scale (or, even small
scale) organizational change, for example. Clearly, the intent of any change effort is to create some emergent macro-
structure with an assumed improved adaptive response to the environment. Well-constructed agent-based models may
provide a valuable platform for examining the patterns in structural emergence and destruction that may occur during
such change efforts. Analysts may even gain insights as to when levels of risk and uncertainty increase during the
evolving process of combating terrorism.
ABM is a platform for exploring many of the phenomena of importance to the process of discovery in public ad-
ministration. ABM provides an opportunity to better understand how structures of importance to governance, from
coalitions to bureaucracy itself emerge. The notion of a generative public administration is also of importance. Public
administration scholars conduct historical analyses of the creation and maintenance of governmental structure, bu-
reaucracy and even the rules that guide bureaucracy. However, generative public administration may provide further
insights into how decentralized heterogeneous agents produce cascades of events that lead to emergent structures.
Researchers may have at their means an improved method for also determining the interactions and intersections of
how agent ‘‘rules’’ guide the evolution of various types of governmental structure.
ABM may also serve as a useful guide in the worlds of applied policy analysis and policy design. Holland [15] has
posited the notion of ABM serving as policy flight simulators. Agent-based models could be used to provide analysts
and policymakers with a variety of alternative scenarios in hopes of finding improved alternatives or the outcomes
of intended policy interventions. This form of rationality, naturally does not overcome the more general problem of
‘‘speaking truth to power’’ but may serve to, at a minimum, improve the level of debate and increase or limit the set of
perceived reasonable alternatives.
6. Conclusion
In an incisive and important work, the anthropologist, Joseph Tainter [16], has examined the decline of several
historic cultures considered highly complex for their particular historical era. This chief conclusion was that their in-
ternal complexity dampened the effect of even their most robust policy efforts. The available set of solutions to such
complex societies is limited in its efforts. The marginal returns of each new innovation became smaller and smaller.
Eventually, there is no margin for error, and a shock to the system may yield disastrous system level analysis. Almost
surely efforts to understand the collapse of the Soviet Union in the late 1980s may benefit from efforts to model political
processes as a phenomenon of adaptive complexity.
We know that as systems become more highly dependent, a change in any element of the system can create a
‘‘cascade’’ in which all elements in the system are affected. In the former Soviet Union, the reality of a planned economy
meant that the failure of planners to anticipate correctly the demand for light-bulbs in Smolensk could create a ripple
effect that had an impact on various sectors; and, given the inextricable interweaving of the economic and political
systems, failure in the economic sphere inextricably created tensions and, ultimately, a dangerous loss of political le-
gitimacy.
Similarly, an understanding of CAS and new methodologies for modeling nonlinear adaptive structures such as
ABM may allow us to gain insights into how democratic societies can develop organizations and institutional
frameworks that allow them to first, fend off terrorist attacks similar to those that occurred in the recent past, and
second, when such attacks do occur, how to assume that the nation possesses a sufficient degree of organizational
flexibility and adaptability so that such shocks do not yield outcomes that massively disrupt our political, economic and
social life.
References
[1] Dooley K. Available from: http://www.eas.asu.edu/~kdooley.
[2] Mol A, Law J. Regions, networks and fluids: anaemia and social topology. Social Stud Sci 1994;24:641–71.
68 E. Elliott, L.D. Kiel / Chaos, Solitons and Fractals 20 (2004) 63–68
[3] Mol A, Law J. Regions, networks and fluids: anaemia and social topology. Social Stud Sci 1994;24:643.
[4] Gilbert N, Doran J, editors. Simulating societies: the computer simulation of social phenomena. London: UCL Press; 1994.
[5] Casti J. Would-be worlds: how simulation is changing the frontiers of science. New York: John Wiley & Sons, Inc; 1997.
[6] Topa G. Social interactions, local spillovers and unemployment. Manuscript, Department of Economics, New York University,
1997.
[7] Patrick S, Dorman PM, Marsh RL. Simulating correctional disturbances: the application of organization control theory to
correctional organizations via computer simulation. J Artif Soc Social Simul 1999;2(1):1–21.
[8] Prietula MJ, Carley KM, Gasser L. Simulating organizations: computational models of institutions and groups. Cambridge, MA:
MIT Press; 1998.
[9] Cederman L. Emergent actors in world politics: how states and nations develop and dissolve. Princeton, NJ: Princeton University
Press; 1997.
[10] Berry BJL, Kiel LD, Elliott E. Sackler colloquium on adaptive agents, intelligence, and emergent human organization: capturing
complexity though agent-based modeling. Proc Natl Acad Sci 2002;99:7187–316.
[11] Epstein JM, Axtell R. Growing artificial societies: social science from the bottom up. Washington, DC: Brookings Institution
Press; 1996.
[12] Langton CG. Artificial life. In: Langton CG, editor. SFI studies in the sciences of complexity. Addison-Wesley: Redwood City,
CA; 1989. p. 1–47.
[13] Epstein JM. Agent-based computational models and generative social science. Complexity 1999;4(5):41–60.
[14] Epstein JM. Agent-based computational models and generative social science. Complexity 1999;4(5):47.
[15] Holland JH. Hidden order: how adaptation builds complexity. Reading, MA: Helix Books; 1995.
[16] Tainter R. The collapse of complex societies. Cambridge: Cambridge University Press; 1988.