Vision & Motion in Collective Behavior

Abstract

All social interactions are mediated by sensory mechanisms. Ants communicate via pheromone trails. Honeybees communicate with dances involving both motion and sound. Fish in schools and birds in flocks use vision to know where their neighbors are going. Yet our models of collective motion are, in general, non-mechanistic. While simple models have helped us to understand the key ingredients sufficient for qualitatively capturing the complex patterns of behavior we see in the natural world, we cannot use non-mechanistic models to ask simple questions about how those patterns may change when properties of the sensory environment change. What sensory information do individuals even use when they need to decide, in an instant, whether or not they should take energetically-expensive evasive maneuvers, especially when only one among many of their neighbors has just moved as if in imminent danger? What happens when the sensory environment is degraded, and it becomes difficult to detect far-away individuals? Are the group level patterns in this case simply a version of what we might expect from the standard models of collective motion, or do new dynamics emerge? How do we tackle the problem of finding the sensory features—not just the sensory modality—at play in collective behavior? Using a species of schooling fish as a model system, we show that visual information is key to understanding both behavioral cascades and between-group interactions. Simple visual properties of near neighbors predict who will respond to whom when fright waves cascade through the group. In cloudy water, between-group interactions are not at all what would be expected from the typical models of individual-level interactions, but can be explained from a first principles account of the relevant factors of the visual system. Lastly, we demonstrate a general, principled method based on rate-distortion theory, the branch of information theory that deals with compressed representation, which not only provides us with simple, interpretable categorizations of the typical behavior and visual information of schooling fish, but the link between them. In sum, we show both how to uncover the sensory basis of social interactions, and its importance to collective motion in animal groups.