Sunday, June 24, 2012

What I'm Reading, Part One: Complex Adaptive Systems Theory

An article was posted recently on the Social Science Research Network that employed a theoretical model with which I was totally unfamiliar - Complex Adaptive Systems Theory. Many of the leading theorists of this model are out of the Sante Fe Institute, including Fabrizio Lillo, Murray Gell-Mann, Mark Pagel, and John Holland (who coined the term). At the bottom of the page I have included a list of other centers and institutes that are related to this work - many of them offer open-access papers.

The article that sparked my interested is called "What Can Neuroscience Tell Us About Religious Consciousness? A Complex Adaptive Systems Framework for Understanding the Religious Brain," by Aaron Burgess.

So I looked it up in Google and found this resource: the Complex Adaptive Systems Group. From there I downloaded some working papers from the Sante Fe Institute, some more papers from Google Scholar, and then began to read (still only scratching the surface, I got distracted with another theory - see Part Two of this post).

From the CAS-wiki:
A Complex Adaptive System (CAS) is a special case of a complex system which is also adaptive, i.e. it has the ability to change and adapt itself to the environment. Typically it consists of a large number of interacting adaptive agents. CASs are used to understand events, objects, and processes in their relationship with each other. They are 'complex' - they are diverse and made up of multiple interconnected elements - and 'adaptive' - they have the capacity to change and learn from experience. They are systems where a lot of individual adaptive agents interact and communicate with each other: complex MAS with adaptive agents. The agents of a CAS learn and change as they interact with each other. The name complex adaptive systems has been coined at the interdisciplinary Santa Fe Institute (SFI), by John H. Holland, Murray Gell-Mann and others. John H. Holland is one of the inventors of evolutionary computation and genetic algorithms, Nobel Prize laureate Murray Gell-Mann discovered quarks.
I think you can see why someone interested in integral theory might also be interested in this approach to systems (which also includes, as a tangent of sorts, emergence theory). One of the cool things about this model is that it is very scalable, so that it is useful at the molecular levels of disease (for example), the neural level of brain circuits, the etiology of mental illness, the societal level of politics or economics, and the macro level of weather and ecosystems.

When we add in various forms of emergence theory, especially when it's informed by causal pluralism (more on this in another post), the possibilities are vast. This is my kind of rabbit hole.

Also from the wiki, here is a little more on the properties of CAS:

As Robert Eisenstein, the former president of the SFI said in the SFI Bulletin from Winter 2004 Vol. 19 No. 1, "despite differences in substrate, there are common principles and mechanisms that underlie the processes by which nature organizes complex systems and how they behave. In other words, there is often simplicity within complexity." These common principles are for example universal scaling laws, similar underlying small-world or scale-free networks in complex networks, or similar forms of emergence, co-evolution and self-organization.

Simon A. Levin (2002) mentions diversity, localization and autonomy as the essential properties of a CAS:
  • diversity and individuality of components,
  • localized interactions among those components, and
  • an autonomous process that uses the outcomes of those interactions to select a subset of those components for replication or enhancement.
According to Stephanie Forrest (1994), the term CAS refers to a system with the following properties:
  • Multi-Agent System A collection of primitive components, called "agents"
  • INTERACTION Interactions among agents and between agents and their environments
  • Emergence Unanticipated global properties often result from the interactions
  • Adaptation Agents adapt their behavior to other agents and environmental constraints
  • Evolution As a consequence, system behavior evolves over time
The essential property that distinguishes a complex adaptive system from a merely complex one is certainly adaptation. Adaptive means agents or populations of agents are able to learn. Either individual agents learn or the population learns, i.e. the population is subject to a process of mutation and competitive selection. In any case, learning agents are able to modify their rules according to their previous success in reacting to the environment. The more successful rules are selected, whereas the less successful rules are deleted.
Wikipedia also has an entry on this model - it offers the following general properties and sets of characteristics from two different versions of the model:

General properties

What distinguishes a CAS from a pure multi-agent system (MAS) is the focus on top-level properties and features like self-similarity, complexity, emergence and self-organization. A MAS is simply defined as a system composed of multiple interacting agents. In CASs, the agents as well as the system are adaptive: the system is self-similar. A CAS is a complex, self-similar collectivity of interacting adaptive agents. Complex Adaptive Systems are characterised by a high degree of adaptive capacity, giving them resilience in the face of perturbation.
Other important properties are adaptation (or homeostasis), communication, cooperation, specialization, spatial and temporal organization, and of course reproduction. They can be found on all levels: cells specialize, adapt and reproduce themselves just like larger organisms do. Communication and cooperation take place on all levels, from the agent to the system level. The forces driving co-operation between agents in such a system can, in some cases be analysed with game theory.


Complex adaptive systems are characterized as follows[3] and the most important are:
  • The number of elements is sufficiently large that conventional descriptions (e.g. a system of differential equations) are not only impractical, but cease to assist in understanding the system, the elements also have to interact and the interaction must be dynamic. Interactions can be physical or involve the exchange of information.
  • Such interactions are rich, i.e. any element in the system is affected by and affects several other systems.
  • The interactions are non-linear which means that small causes can have large results.
  • Interactions are primarily but not exclusively with immediate neighbours and the nature of the influence is modulated.
  • Any interaction can feed back onto itself directly or after a number of intervening stages, such feedback can vary in quality. This is known as recurrency.
  • Such systems are open and it may be difficult or impossible to define system boundaries
  • Complex systems operate under far from equilibrium conditions, there has to be a constant flow of energy to maintain the organization of the system
  • All complex systems have a history, they evolve and their past is co-responsible for their present behaviour
  • Elements in the system are ignorant of the behaviour of the system as a whole, responding only to what is available to it locally
Axelrod & Cohen[4] identify a series of key terms from a modeling perspective:
  • Strategy, a conditional action pattern that indicates what to do in which circumstances
  • Artifact, a material resource that has definite location and can respond to the action of agents
  • Agent, a collection of properties, strategies & capabilities for interacting with artifacts & other agents
  • Population, a collection of agents, or, in some situations, collections of strategies
  • System, a larger collection, including one or more populations of agents and possibly also artifacts.
  • Type, all the agents (or strategies) in a population that have some characteristic in common
  • Variety, the diversity of types within a population or system
  • Interaction pattern, the recurring regularities of contact among types within a system
  • Space (physical), location in geographical space & time of agents and artifacts
  • Space (conceptual), “location” in a set of categories structured so that “nearby” agents will tend to interact
  • Selection, processes that lead to an increase or decrease in the frequency of various types of agent or strategies
  • Success criteria or performance measures, a “score” used by an agent or designer in attributing credit in the selection of relatively successful (or unsuccessful) strategies or agents.
Looks cool, eh?

Here are some of the articles I was able to find on this model (or related to it) that are freely available online (in no particular order):

Related Papers:
Groups and Institutes. International research groups and institutes exploring complex systems:
  • SFI Santa Fe Institute
  • CSCS Center for the Study of Complex Systems
  • CCSR Center for Complex Systems Research
  • CASOS Center for Comp. Analysis of Social and Org. Systems
  • ICES Institute for Complex Engineered Systems
  • NECSI New England Complex Systems Institute
  • NICO Northwestern Institute on Complex Systems
  • ECCO Evolution, Complexity and Cognition group

Post a Comment