// The Song Remains The Same

(This post is an introduction to the basis ideas of self organization. Over the coming weeks I’m going to take a look at the various aspects of self organization in living and non living systems, and how that is being applied to the computational and linked data worlds.)

Technological sytems become organized by commands from outside, as when human intentions lead to the building of structures or machines. But many natural systems become structured by their own internal processes: these are the self-organizing systems, and the emergence of order within them is a complex phenomenom that intrigues scientists from all disciplines.

— F. E. Yates et al., Self Organizing Systems: The Emergence of Order

The patterns of self organization are prevalent across the natural world, driving all living processes on some level. Systems such as ant colonies, slime molds, tissue formation — all have no centralized control, blueprints, or coordinators. The book “Self Organization in Biological Systems” (Camazine, Deneubourg, et al) describes self organization as:

Self-organization refers to a broad range of pattern-formation processes in both physical and biological systems, such as sand grains assembling into rippled dunes, chemical reactants forming swirling spirals, cells making up highly structured tissues, and fish joining together in schools. A basic feature of these diverse systems is the means by which they acquire their order and structure. In self organizing systems, pattern formations occurs through interactions internal to the system, without intervention by external directing influences.

Self organizing systems exhibit emergent properties, where emergence refers to a process by which a system of interacting agents or nodes acquires qualitatively new properties that cannot be understood as the simple addition of their individual contributions. These agents interact purely at the local level, having no global information. The term emergent property may imply to some an ephemeral property that materializes magically since these system level properties arise unexpectedly from nonlinear interactions among a system’s components.

However, this is not the case given studies such as bark beetle larvae sorting that show how interactive components in nature demonstrate these processes and dispel the myths. No central leader or boss is present to oversee the work, no external orders are given but the process itself continues to function. Self organizing systems such as ant colonies have no leader and rely simply on each ant to execute its job based on only the stimulus from its local environment. With simple rules executed repeatedly based on only the local environment as stimulus, we can observe more complex global phenomenon.

Some alternatives to pattern formation have order imposed on them through techniques such as leaders, blueprints, recipes, or pre-existing patterns in the environment (templates).

• The well informed leader directs the group with instructions about what to do next at each step.
• Another example is the blueprint; it is delivered to a construction crew giving highly detailed descriptions of the complete project which allows them to execute their tasks from.
• A recipe is a set of sequential instructions that precisely specify the spatial and temporal actions of the individual’s contribution to the whole pattern.
• The last mentioned alternative technique is use of a template — a full size guide or mold that specifies the final pattern and strongly steers the pattern formation process.

These techniques are well understood and accepted; However, they should not be confused for the principles of self organization. There are some distinct drawbacks for these techniques when applied to some of the very complex and distributed situations found in nature. For starters, a central authority has a hard time scaling its communication overhead to direct a large number of subordinate units as that number becomes large (and I use the term “large” here in an algorithmic sense). A leader would have to maintain some sort of state for each unit in terms of how its work interacts with all other surrounding units, which doesn’t scale well either. These things place formidable, if not impossible, burdens of information acquisition, processing, and transmission on the leader, especially as the size of the group becomes large and the pattern being built is far larger than any one individual.

The problem with blueprints in biological systems is that it is generally extremely costly to genetically encode the vast quantity of information in a mental blueprint that would be needed to build a termite nest, let alone taking into account the fact that a single termite would not have the cognitive abilities to execute the blueprint. Recipes tend to break down when executed in uncertain environments due to their inability to adapt, and nature is full of uncertain situations. Lastly, templates can show certain patterns in limited situations, but they simply are not available in certain environments. However, self organizational systems still manage to thrive in these uncertain environments.

So how exactly would such a process work if it doesn’t use more common techniques such as a hierarchical command structure or a blueprint? Why do biological systems often rely on interactions among their components rather than instructions from a leader or a external direct instruction source?

A reasonable suggestion is that pattern-formation by cooperative groups usually arises through self organization rather than external guidance because the latter mechanisms generally are not computationally cost effective to execute or too hard to implement. Pattern formation through groups of large agents or subunits through self organization is based on rather simple instructions, which we perceive as rules of behavior, that are easily implemented by each member in the group.

Two core principles that drive and control self organizing processes and will be discussed in coming articles. These principles are feedback (both positive and negative), and stigmergy. Feedback is the effect of dampening or amplifying a stimulus, such as the regulation of blood sugar levels in our bodies and the homeostatic regulation of body temperature in warm blooded mammals. With positive and negative feedback, we can create change in a system, which in turn can create patterns under the right conditions. Stigmergy is about how organisms or agents use intermediary work from other agents to adjust their current behavior or switch tasks entirely. Agents can use stigmergy in conjunction with other local environment information to alter their behavior. Together, the mechanics of feedback and stigmery can create some very complex and emergent behaviors.

A lot of research is active in this field, from Marco Dorigo’s work with ant colonies, to John Holland’s (father of genetic algortihms) work in constrained generating procedures, to a good friend of mine’s work (Jesse St Charles, phd student at CMU) in document clustering using flocking techniques on GPUs. For my master’s thesis, I used the self organization based routing algortihm “Termite” to route packets in a wireless ad hoc network. The core principles of self organization spider out into almost every imaginable topic — but how do they connect to the web data ecosystem? That’s something I want to look into over a series of blog posts. With my next article, I talk about the mechanics of “Positive Feedback“.