From Clusters to Coping

A few years ago, a co-worker was nice enough to buy me a copy of Steven Johnson's book "Emergence." It provided a brief readable update of the self-organizing systems literature I had read 5-10 years before.

I recently ran across an article by Johnson discussing two types of emergent dynamics: clustering and coping. Although his discussion is grounded in his involvement in the Howard Dean campaign, the broader points he makes are potentially relevant to any emergent goal-directed system.

He asserts that Clustering is a simple form of emergent behavior where a shared common interest (possibly a goal) and the following of a few simple rules & signals results in complex group behavior. Examples most folks are familiar with are the boids simulation and slime mold. However, this behavior is not very adaptable.

Coping, by contrast, is characterized by (a) a much more sophisticated semantic code, and (b) the availability of "truly local" meta-information about the group's state (often communicated indirectly). He cites E. O. Wilson's estimate that an ant's pheromone signaling "vocabulary" contains as many as two dozen "words." And, the complexity of the individual agent's rules is correspondingly greater than the simpler Clustering agents.

My initial reaction (trying to bend this compare/contrast to the frames I know) was that COP's cluster, while COI's cope. I'm not so sure his taxonomy fits COPs/COIs very well, but I found it interesting that his discussion of social insects (vs., say slime mold) did not extend to whether Clustering might be characterized by a single decision role/responsibility/right (i.e., homogenous agents), and that Coping might require multiple decision roles/responsibilities/rights (i.e., heterogeneous agents...possibly with the ability to "morph" (i.e., we have 200+ types of cells that emerge from the union of two cells)).

Regardless, there seems to be an implication that designing independent agents to exhibit emergent behavior that is both (a) goal-directed and (b) adaptable may require a taxonomy of roles/responsibilities/rights, a non-trivial signaling code, and the communication of state related to the group and its goal. I don't have a clue about how hard this might be to design, but I think I'm safe in saying it's a different way of designing than most of us are familiar with.