User:Fred Drumlevitch/Sandbox

‘’’Task Allocation and partitioning’’’ refers to the way that tasks are subdivided, assigned, and coordinated within a single colony of social insects. Closely associated are issues of communication that enable these actions to occur. This entry focuses exclusively on social insects. For information on human task allocation and partitioning, see Division of labour,Task analysis, and Workflow.

Introduction

Social living provides a multitude of advantages to its practitioners, including predation risk reduction, environmental buffering, food procurement, and possible mating advantages. The most extreme form of sociality is eusociality, characterized by overlapping generations, cooperative care of the young, and reproductive division of labor, which includes sterility or near-sterility of the overwhelming majority of colony members. With few exceptions, the practitioners of eusociality are all insects of the orders ‘’Hymenoptera ‘’(ants, bees, and wasps), ‘’Isoptera’’ (termites), and ‘’Homoptera’’ (aphids).[cite]. Social insects have been extraordinarily successful ecologically and evolutionarily. This success has at its most pronounced produced colonies 1) having a persistence many times the lifespan of most individuals of the colony, and 2) numbering thousands or even millions of individuals. Social insects generally exhibit division of labor with respect to non-reproductive tasks, in addition to the aforementioned reproductive one. In some cases this takes the form of markedly different, alternative morphological development, as in the case of soldier castes in ants and aphids, while in other cases it is age-based, as with honeybee foragers, who are the oldest members of the colony (with the exception of the queen). Division of labor, large colony sizes, temporally-changing colony needs, and the value of adaptability and efficiency under Darwinian competition, all form a theoretical basis favoring the existence of evolved communication in social insects. [cite]. Beyond the rationale, there is well-documented empirical evidence of communication related to tasks; examples include the “pollen dancing” of honeybee foragers, trail marking by ant foragers, and the propagation via pheromones of an alarm state in “Africanized” bees.

Network representation of tasks and communication

Numerous scientists have used a social network approach to model communication in animals, including that related to task performance. O’Donnell has coined the term “worker connectivity” to stand for “communicative interactions that link a colony’s workers in a social network and affect task performance”. A network is pictorially represented as a graph, but can equivalently be represented as an adjacency list or adjacency matrix. Traditionally, workers are the nodes of the graph, but Fewell prefers to make the tasks the nodes, with workers as the links. O’Donnell has pointed out that connectivity provides three adaptive advantages compared to individual direct perception of needs:

1. It increases both the physical and temporal reach of information. With connectivity, information can travel farther and faster, and additionally can persist longer, including both direct persistence (i.e. through pheromones), memory effects, and by initiating a sequence of events.
2. It can help overcome task inertia and burnout, and push workers into performing hazardous tasks. For reasons of indirect fitness, this latter stimulus should not be necessary if all workers in the colony are highly related genetically, but that is not always the case.
3. Key individuals may possess superior knowledge, or have catalytic roles. Examples, respectively, are a sentry who has detected an intruder, or the colony queen.

O’Donnell provides a comprehensive survey with examples of factors that have a large bearing on worker connectivity. They include:

• graph degree
• size of the interacting group, especially if the network has a modular structure
• sender distribution (i.e. a small number of controllers vs. numerous senders)
• strength of the interaction effect, which includes strength of the signal sent, recipient sensitivity, and signal persistence (i.e. pheromone signal vs. sound waves)
• recipient memory, and its decay function
• socially-transmitted inhibitory signals, as not all interactions provide positive stimulus
• specificity of both the signal and recipient response
• signal and sensory modalities, and activity and interaction rates

Task Taxonomy and Complexity

Anderson, Franks, and McShea have broken down insect tasks (and subtasks) into a hierarchical taxonomy. They classify tasks as individual, group, team, or partitioned; classification of a task depends on whether there are multiple vs. individual workers, whether there is division of labor, and whether subtasks are done concurrently or sequentially. Note that in their classification, in order for an action to be considered a task, it must contribute positively to inclusive fitness; if it must be combined with other actions to achieve that goal, it is considered to be a subtask. In their simple model, they award 1, 2, or 3 points to the different tasks and subtasks, depending on its above classification. Summing all tasks and subtasks point values down through all levels of nesting allows any task to be given a score that ranks relative complexity of actions.

Note: Model-building

All models are simplified abstractions of the real-life situation. There exists a basic tradeoff between model precision and parameter precision. A fixed amount of information collected, will, if split amongst the many parameters of an overly precise model, result in at least some of the parameters being represented by inadequate sample sizes. Because of the often limited quantities and limited precision of data from which to calculate parameters values in non-human behavior studies, such models should generally be kept simple. Therefore we generally should not expect models for social insect task partitioning to be as elaborate as human workflow ones, for example.

Other metrics for task partitioning

With increased data, more elaborate metrics for task partitioning become possible. Gorelick and Bertram survey the applicability of metrics taken from a wide range of other fields. They argue that a single output statistic is desirable, to permit comparisons across different population sizes and different numbers of tasks. But they also argue that the input to the function should be a matrix representation (of time spent by each individual on each task), in order to provide the function with better data. They conclude that “…normalized matrix-input generalizations of Shannon’s and Simpson’s index … should be the indices of choice when one wants to simultaneously examine division of labor amongst all individuals in a population”. Note that these indexes, used as metrics of biodiversity, now find a place measuring task partitioning.