The Power of Social Information in Ant-Colony House-Hunting: A Computational Modeling Approach

Abstract

The decentralized cognition of animal groups is both a challenging biological problem and a potential basis for bio-inspired design. The understanding of these systems and their application can benefit from modeling and analysis of the underlying algorithms. In this study, we define a modeling framework that can be used to formally represent all components of such algorithms. As an example application of the framework, we adapt to it the much-studied house-hunting algorithm used by emigrating colonies of Temnothorax ants to reach consensus on a new nest. We provide a Python simulator that encodes accurate individual behavior rules and produces simulated behaviors consistent with empirical observations, on both the individual and group levels. Our model successfully reproduces experimental results showing the high cognitive capacity of colonies, their rational time investment during decision-making, and their ability to avoid and repair splits with the help of social information. We also use the model to make predictions about several unstudied aspects of emigration behavior. The results suggest the value of individual sensitivity to site population for ensuring consensus, and they indicate a more complex relationship between individual behavior and the speed/accuracy trade-off than previously appreciated. The model proved relatively weak at resolving colony divisions among multiple sites, suggesting either limits to the ants’ ability to reach consensus, or an aspect of their behavior not captured in our model. It is our hope that these insights and predictions can inspire further research from both the biology and computer science community.

A Geometry-Sensitive Quorum Sensing Algorithm for the Best-of-N Site Selection Problem

“…To better study the ants' behavior, models have been designed to simulate how ants change behavior throughout the house hunting process [15,24]. These models, initiated by Pratt [15], allow simulated ants to probabilistically transition through four phases -the Exploration, Assessment, Canvassing, and Transport phases.…”

“…This assumes that any nest is equally likely to be found, which is unlikely in the real world because sites closer to the home nest are more likely to be discovered. To our knowledge, house hunting models have not tried to model situations where nests have different likelihoods of being found, as is the case when nests have different distances from the home nest [24].…”