Stigmergy in the generalized reading frame

by Suspended Reason

Stigmergy” is a term from entymology and complexity studies to describe behavior such as termites building colony mounds. It is the idea that each (e.g.) termite’s decision about where and how to contribute to the mound is founded entirely in the existing information encoded in the existing structure-to-date. Since this structure is the product of other termite actions, stigmergy is a general principle whereby agents’ actions leave cues behind which inform other agents’ actions.

I have found most definitions of stigmergy incoherent and difficult to apply, and will try to make it more amenable to TIS frameworks. Its original formulation by Pierre-Paul Grassé (1959) defined it as stimulation of the workers by the performance they have achieved.” This, unfortunately, is synonymous with a definition of cooperative communication, and Grassé’s implicit distinction between the information encoded in the performance that is the termite mound, and linguistic performances, is nowhere made explicit.

Holland and Melhuish (1999) double-down on this confused definition: All that is necessary for stigmergy to occur is for the outcome of the behavior of the relevant agent to be appropriately affected by previous environmental changes.” And yet what is communication except the alteration of other agents’ behaviors via environmental change (in which all information” must necessarily be encoded). Doyle and Marsh (2012) attempt a refactor of stigmergy which exacerbates the problem, emphasizing stigmergy as any use of the environment as a memory or communication medium. I would invite the authors to provide an example of communication between any organisms, agents, or machines that is not, in this factoring, stigmergic.

In the generalized reading framework, all actions produce (“write”) information, whether or not they are performed for the purpose (solely or partially, consciously or unconsciously) of writing. Other agents use this information in their own decision-making as part of normal ecological intelligence and optimization. In this sense stigmergy” seems like normal communication, albeit one scoped to coordinative” (positive-sum) behaviors. It is difficult to see how this might be a usable or coherent concept in its own right.

My proposed amendment or refactoring is to define stigmergy as a quality or spectrum, with its ideal or limit case, as a cooperative relationship in which all the information used in decision-making is a byproduct of actions geared toward intrinsic ends. There is no communicative surplus” in the sense of cooperator actions whose sole or primary purpose is the alteration of other cooperator’s actions (all language being such surplus). Thus human activities may be more or less stigmergic with respect to how much surplus information (be it verbal or non-verbal) is produced by cooperating agents in the achievement of some project.

Taking Grassé’s intension seriously, a traffic light is stigmergic, since its existence is the result of a performance by government workers (members of the larger population) which alters the behavior of other workers (those driving, walking, etc). But ameliorating this intension to our own—which I believe better captures Grassé’s intended extension—the traffic light has been erected solely for the purpose of communicating information, and for no other purpose. This is a wholly different sort of situation than pheromone trails in ants, where as a byproduct of normal movement, ants leave behind a traceable set of decentralized trails for other workers. All the information necessary or provided to cooperators is encoded in normal,” extrinsic, non-communicative” action.