Colony of marked L. curvispinosus

Tandem run

Video of tandem run

Transport

Video of transport

Collective nest site choice by ant colonies

My work is focused principally on a striking example of collective decision-making by ants of the genus Temnothorax. These ants have special advantages as a model system for analyzing collective behavior. They form colonies small enough that all workers can be individually marked, and they thrive in glass-walled artificial nests that facilitate detailed video image analysis of every relevant social interaction. In nature they frequently emigrate when their old nest deteriorates, and migrations can be induced in the laboratory simply by removing the upper slide from the nest and providing intact new nests nearby. Given a choice between nests of different sizes, colonies not only avoid dividing among more than one nest, but also consistently choose a new home of a preferred design (Mallon et al. 2001; Pratt and Pierce 2001). This is no small feat, given that each nest is evaluated by only a few ants, who do not have the opportunity to directly compare all available sites.

We have found that the colony's distributed decision-making ability relies on a novel multi-stage decision process used by the minority of active ants who organize the move (Mallon et al. 2001; Franks et al. 2002; Pratt et al. 2002). An active ant initiates recruitment to a promising site only after a delay that varies inversely with site quality. Recruitment creates positive feedback on the number of ants visiting the site, but the quality-dependent delay ensures that feedback is stronger to better nests. The ants amplify this difference by using two distinct forms of recruitment: a slow method (tandem running) for their fellow active ants, and a faster method (transport) for the passive ants and brood that make up the bulk of the colony. The ants first use the slow method, switching to rapid transport only after the new site has reached a threshold population. A differential equation model has shown that this quorum rule raises the accuracy of the colony's decision by reducing the likelihood of carrying passive ants to an inferior site.

In collaboration with David Sumpter of Oxford University, I am currently developing an agent-based computer simulation of the nest-choice process, incorporating everything we have learned about individual behavior. The thorough data collection and high replication rate attainable with this system allow us to make reliable estimates of all model parameters, including probabilities of inititating recruitment to a nest of given quality, or the quorum size demanded before transport begins. We have successfully validated the model, finding good agreement between predicted group-level performance and that seen in real colonies. We can now use it to probe more deeply into the colony's decision mechanisms, both by conducting virtual experiments and by testing its predictions in real experiments. For example, we will derive simpler models from the full one, to specify the aspects of individual behavior that are necessary for successful group-level behavior. We will explore the sensitivity of the decision algorithm to variations in parameter values, and the possible role played by variation among individual ants. We will test the robustness of the algorithm by examining its performance in a variety of environmental contexts, such as different numbers and arrangements of new nests, and different degrees of urgency to move. The latter will allow us to determine how ants deal with possible trade-offs between speed and accuracy of decision-making, depending on the severity of conditions at the old nest.

Recent experiments are also beginning to reveal how individual ants detect the presence of a quorum. They suggest that the crucial cues come from direct encounters with other ants in the new nest, integrated over time to give each ant a measure of her encounter rate with nestmates. This mechanism shows interesting analogies to models of categorical decision-making in humans, in which sensory evidence drives a random-walk from a neutral initial value toward either a high or a low threshold, the first threshold to be reached determining which option is chosen. A key feature of these models is that they show how the trade-off between speed and accuracy of decision-making can be regulated, simply by moving the thresholds farther from or closer to the starting point. In the case of individual human decision-makers, the brain conceivably optimizes this adjustment by linking it to the individual's rate of reward. In the case of ants judging nest populations, a similar model may apply, except that random encounters with nestmates supply the crucial sensory input. In the ants' case, optimal decision thresholds will be determined not by their effect on current individual reward, but by their effect on the fitness returns to the colony as a whole, measured in terms such as duration of emigration, quality of nest eventually chosen, or risk of colony splitting. Thus, by bringing our analysis of decision-making to a finer level of resolution, we will be able to ask more precisely how individual cognitive abilities are designed to optimize group behavior.