Laboratory of Intelligent Systems - Division of labour in social insects



Evolution of Division of Labour in Social Insects



Understanding division of labour
in social insects using agent
based models and artificial evolution.


Danesh Tarapore,
Steffen Wischmann,
Prof. Dario Floreano,
Prof. Laurent Keller



A leafcutter ant.
(c) www.forestryimages.com
Social insects are known for their high degree of cooperation and complex social organisation that have enabled them to form colonies with millions of individuals and underground cities with tunnel networks spanning many kilometres. One aspect that has long fascinated biologists and engineers alike are the intricate and often highly specialized mechanism of division of labour between different castes or age groups observed in many species: A mature leafcutter ant colony, for example, typically consists of more than 2 million individuals, which are divided into morphologically specialized castes: Some workers exclusively nurse the brood, others forage for food and still others take care of fungi gardens. A specialized soldier caste defends the colony from intruders. Similar phenomena can be found in many other social insects. While honey bees, for instance, do not show morphological specialisation, it has been shown that individual bees dynamically adapt their foraging behaviour of pollen and nectar as well as water as a function of individual preference and colony need.

Unlike centralized hierarchical systems of division of labour typical of human societies, social insects exhibit a completely decentralized system of task allocation, the division of labour resulting from interactions between members of the colony and between the colony and the environment. While this leads to highly complex dynamics governed by many independent individual interactions, this decentralized system has many advantages when compared to a hierarchical organisation: Scalability, robustness, flexibility and simplicity. Scalability in terms of the number of individuals in the colony, robustness and flexibility in terms of the colony coping with environmental perturbations and simplicity in terms of the behaviour of each individual of the colony. These properties have led to an increased interest in the dynamics and organisation of social insect colonies in many domains outside of biology, including network routing, optimization theory and robotics.

Worker polymorphism in the mound ant Formica obscuripes. These ants are sisters from the same nest.
(c) Alex Wild 2004, www.myrmecos.net


Biologists have tried to gain a detailed understanding of the division of labour in social insect colonies and the evolution of social organisation using a variety of methods. On the one hand, theoretical work such as kin selection theory or reproductive skew models have greatly contributed to our understanding. While such theories are often ill suited to predict individual dynamics or point out the mechanisms involved in individual interactions, they excel at making general predictions that can be verified at the colony and population level. Many experimental approaches, on the other hand, focus on the genetics of social insects.

Major and minor green tree ants standing atop their silken arboreal nest.
(c) Alex Wild 2004, www.myrmecos.net
In recent years, a third approach to unravelling the dynamics of social insect colonies has gained increasing popularity. Agent based computer models can make a significant contribution by combining theoretical results applicable at the colony level with individual task choice. This allows to directly analyse the effects of changes at the genetic level at the colony level. A number of factors including colony composition, division of labour and reproductive skew can be analysed in this context.

It is important to note that while such computer models are a useful tool to quickly test hypothesis for their plausibility and can generate new insight into colony system dynamics, models cannot by themselves serve as a proof for biological mechanisms or replace biological experiments. Ideally, they generate predictions about the biological system and lead to the development of empirical tests.

In addition to their interest for biologists, abstract models for division of labour and task allocation are also of great interest to other agent based systems, such as multi-robot teams. While many robot tasks can be solved using hierarchical, modular systems, in some cases the advantages of so-called robot swarms outweigh added difficulties in design and trouble shooting. This is especially true for problems involving large numbers of low cost robots.

In this project we are interested in the influence of different genetic architectures on division of labour and overall colony task performance in both, the biological system and their potential for controlling swarms of robots.


Effect of genetic architecture and polyandry on division of labour in social insects.

Genetic diversity is thought to be a main factor in determining task performance and behavioural plasticity of social insect colonies. This diversity has two main causes. a) Multiple matings by the colony queen (polyandry) and b) the number of regions on the genome that influence a behavioural trait (loci). However experiments exploring the relation between these two factors in influencing division of labour are relatively rare due to the difficulties associated with performing them.

Visualization of the experiment setup consisting of workers foraging for circular and rectangular objects, bringing them back to the nest.
In this project we simulate the evolution of a colony of workers which are offspring of a single queen mating with a variable number of drones. Simulations have been carried out using a probabilistic agent based simulator, while varying the amount of polyandry and the number of loci encoding for the two response thresholds. A colony is evaluated based on the ability of its workers to bring resources of two types to the nest. While the workers have to collect as many circular resources, the number of rectangular resources in the nest have to be maintained within an upper and lower bound. This takes inspiration from nature, an example being the diet of a carpenter ant colony, which includes honeydew from aphids, sweets, meat, and fats.


Example of the genetic architecture of a worker. The alleles in the yellow and blue regions encode for the first and second response threshold respectively.
The colony evaluation motivates the workers to switch between the two foraging behaviours depending on the state of the colony. A workers decides on its foraging behaviour using the response threshold model, the response thresholds for the two foraging behaviours being encoded in the worker's genome.

Interestingly we find that the colony phenotypic diversity and foraging efficiency improves significantly with an increment in polyandry. We also find that when using an additive model of gene expression, an increment in the number of loci influencing a foraging trait causes a decrement in the phenotypic diversity and colony fitness.

This experiment provides interesting insights into division of labour in social insects. It may also prove useful in the field of robotics, when multi-agent systems are needed to solve a problem, the agents being the workers in our experiment. The colony structure allows for the genetic information of all the agents to be contained in the queen and drones of the colony. During evolution, the genetic operators are applied on the few genomes of the queen and drones instead of being applied individually on the large number of workers.

The recombination genetic operator is applied across the few genomes of a queen and drones to produce a large number of workers for the next generation of evolution. Without our colony structure, recombination would have to be applied across the large number of individual workers of different colonies, requiring a very large number of colonies in the population.


Publications

301 Moved Permanently

Moved Permanently

The document has moved here.



People actively working on the project


Previous projects and collaborators

Previous collaborators Previous student projects
Last update: 08.03.09/mw