Consensus Decisions and Structural Information in Shoaling Fish

Abstract

Non-hierarchical animal groups, such as bird flocks or fish schools, are unique computational entities. They have the ability to make collective decisions rapidly and robustly, mediate conflict, and navigate complex environments. While these collective computations are reminiscent of human voting or neuronal systems, schools and flocks achieve these ends without the benefit of any aggregation layer or hierarchical control. There are no clerks assigned to tally votes nor are there any special neurons which sum, average, or compute other functions based on the activity of neurons further upstream. Instead, certain computations are emergent in nature; local interactions lead to extraordinary self-organized behaviors. In this way, flocks and schools closely resemble ferromagnetic physical systems.

In this work, we investigate collective computation in the context of golden shiners (Notemigonus crysoleucas), a species of strongly schooling minnow that are ubiquitous in freshwater habitats across North America. First, we investigate how, and under what conditions, these schools are able to form consensus decisions amidst internal conflict regarding the direction of travel. Specifically, we examine an asymmetric scenario where a strongly preferenced minority is in conflict with a more weakly preferenced majority. We show that the resolution of this conflict depends on uninformed individuals or those with no stake or even awareness of the conflict. Next, we recast this biological problem as a well-studied spin system. We provide a theoretical framework for understanding the central role of uninformed individuals and demonstrate that the direction of their influence is dependent upon the linearity or non-linearity of local interactions.

Lastly, we investigate local structure within schools. Because school-level behaviors emerge from the self-organization of local interactions, local neighborhoods are of particular interest and importance. Employing techniques from statistical mechanics and computational neuroscience, we build precise generative models of local configurations from both perturbed schools in a narrow channel and schools free to explore the full experimental tank. These models, while simpler than complex sensory models, allow us to develop a statistical mechanics for school structure and to investigate how, and to what degree, local structure predicts future individual and group behavior.