Democracy and the Emergence of Coexistent Communities
This paper asks the question: Do the institutions
and processes of democracy provide sufficient conditions to enable the emergence
and persistence of coexistence among diverse human groups? A powerful
and persuasive social analysis has resulted from the pioneering work of R.
J. Rummel, who documents that well-established democratic societies do not
go to war against one another and have less internal violence. Explanations
for “the democratic peace” generally fall into propositions about the adaptive
nature of democratic governance structures and the enabling nature of democratic
culture. Democratic structures manifest two collective rules:
freedom of choice (including choice of leaders) and the rule of law.
Democratic culture manifests rules that impact on the individual level:
tolerance of diversity and willingness to accommodate competing demands through
the exchange of resources. While we believe the reasoning and conclusions
are correct, we do not believe sufficient understanding exists of the emergent
relationship between individual behavior and global expression, nor of the
dynamics of interactions within and between different governance structures.
To future this understanding, we design a suggestive
agent-modeling simulation that allows us to compare different political communities
as examined by Rummel: an “autocratic” community, a totalitarian community
and a “democratic” community. The agent system (a variation of a consumer
simulation) is based on a model of individual behavior that captures dynamic
excursions from habitual behavior: the cognitive, social and authoritative
decision making (based on an extension of the work by Janssen and Jager),
corresponding to a rational agent, a social agent and a deferential agent.
This individual-based agent model is argued to capture the three different
forms of governance depending on the collective state of the individuals
in a dynamic, path-dependent process.
Finally, the degree of (political) coexistence within
a community and likelihood of non-violence between communities is examined
as a function of rates of change of resource availability and location (argued
to be a source of stress on the system). The behaviors of the model
are suggestive of the conclusions of Rummel.
Merle Lefkoff
Los Alamos National Laboratory
Center for Nonlinear Studies
Norman L. Johnson
Los Alamos National Laboratory
Theoretical Division
Issues in using agent-based simulations for prediction
(rather than description)
An agent-based simulation, almost by definition, represents
a single realization out of many possible realizations. One way to demonstrate
this is by repeating a simulation with subtle changes either to the initial
conditions, the boundary parameters, or the model choices. The resulting
simulation will typically give a final state that is different in a noticeable
way, thereby representing a different "realization" of a state of the system.
Most agent simulations easily capture this chaotic or nonlinear sensitivity
to small changes in the simulation. This chaotic behavior of the simulations
leads to uncertainty in the uniqueness of the simulation as a description
of a collection of events. Consequently, the utility of most agent
simulations is obtained either by repeating simulations to obtain a stochastic
characterization of the system behavior or by carefully tuning the initial
and operating parameter to duplicate a desired outcome. These uses
of agent models are largely descriptive, attempting to provide understanding
of the connection of agent behavior and interactions to global behavior (the
self-organization or emergence problem). In the near future, there
will be an alternative need of agent simulations: to predict future states
from knowledge of past states. This is comparable to the use of weather
simulations tools to predict the tomorrow's weather based on yesterday's
data. Similarly, large numbers of rich social behavior agent-based
simulations will be asked to predict future outcomes from known histories.
This will be particularly true as rich data sources are developed that are
compatible with agent simulations, for example the prior purchasing histories
of a group of consumers or the history of an epidemic. This predictive
use of agent simulations is equivalent to recasting the unconstrained simulations
with internal constraints that match the known data. The current
work demonstrates the challenges associated with such a predictive use of
agent simulations and suggests possible general approaches to the problem.
The analysis illustrates how the assumptions of classical methods of predicting
future states of distributed systems, e.g., statistical mechanics, are violated
by the network nature of agent model, in which the probability distribution
functions are highly localized rather than being applicable to the entire
system. This local isolation of the evolution of the "phase space"
or operational space is found to duplicate new understandings in the fields
of anomalous distributions and non-extensive thermodynamics.
Norman L Johnson
Los Alamos National Laboratory
http://ishi.lanl.gov
nlj@lanl.gov