A parallel distributed model of the behaviour of ant colonies

Abstract

Members of an ant colony perform a variety of tasks outside the nest, such as foraging and nest maintenance work. The number of ants actively performing each task changes, because workers switch from one task to another and because workers are sometimes active, sometimes inactive. In field experiments with harvester ants (Gordon, 1986, Anita. Behao. 34, 1402-1419, 1987, a perturbation that directly affects only the number of workers engaged in one task, causes changes in the numbers engaged in other activities. These dynamics must be the outcome of interactions among individuals; an ant cannot be expected to assess and respond to colony-level changes of behaviour. Here we present a parallel distributed model of the processes regulating changes in numbers of workers engaged in various tasks. The model is based on a Hopfield net, but differs from conventional Hopfield models in that when a unit or ant changes state, it changes its interaction patterns. Simulation results resemble experimental results; perturbations of one activity propagate to others. Depending on the pattern of interactions among worker groups, the distribution of active workers in different tasks either settles into a single, global attractor, or shows the dynamics associated with a landscape containing multiple attractors.

Fixed Response Thresholds and the Regulation of Division of Labor in Insect Societies

“…The magnitude of task-associated stimuli are not taken into account, which makes the functioning of the system rely solely on an internal dynamics. Gordon et al (1992) were able to reproduce some experimental results qualitatively, but other results are unlikely to be explained, as the model cannot be robust with respect to certain perturbations (for example, because it is insensitive to colony needs). Pacala et al (1996) have recently developed a rather general, simple model of task allocation.…”

“…In an attempt to account for the field observations of Gordon (1986,1987,1989) on the dynamics of task allocation in harvester ants, Gordon et al (1992) have developed a model based on a connectionist (neural net) model, which can be seen as a more complex (and a more experiment-driven) version of Page and Mitchell's (1990) boolean network, where units correspond to ants, possible states of a given unit to tasks that can be performed by the ant, and patterns of connections to observed patterns of interaction among individuals. They modeled eight categories of ants.…”

The organization of work in social insect colonies

“…Nest mound patrollers may assess humidity and temperature. After the nest mound patrollers have gone back in, trail patrollers choose foraging directions, based on encounters with the foragers with neighboring colonies and perhaps on food availability [14,15]. Once foraging has begun, a forager's decision whether to go out to forage depends in part on its rate of contact with successful returning foragers [11,13].…”

“…Colonies live 15 to 20 years, founded by a single queen who produces all the workers. Colonies begin with 0 ants and grow to a size of about 10,000 ants when the colony is about 5 years old and begins to reproduce [14]; it then stays at about this size for another 10 to 15 years. The behavior of older, larger colonies, of about 10,000 ants, is more stable to perturbation and more homeostatic than that of younger, smaller ones of about 3,000 ants [3].…”

“…The first [14] involves a parallel distributed process, such as a neural network. In this model, individual decisions are based wholly on interactions with nestmates.…”