Research‎ > ‎

Information transfer in fish schools

With: Colin Twomey, Nikolai Bode, Albert Kao, Yael Katz, Christos Ioannou, Brin Rosenthal, Colin Torney, Haishan Wu, Simon Levin, and Iain Couzin 

One benefit of living in a group is the ability to make use of distributed sensing capabilities, which may aid in both predator avoidance and resource acquisition. In order for individuals to take advantage of these capabilities, information must be reliably transmitted across the group. Unraveling the process of information transfer is key to our understanding of collective behavior. We are using fish schools as a model system to explore how information is transmitted across groups.

One way to probe this process is by introducing a perturbation into a group and tracking its spread. Here, a small proportion of “informed” fish, trained to respond to a stimulus, were introduced into an uninformed group. This setup allows us to test theoretical predictions about how the ability of informed fish to lead the group scales with their number. Moreover, by using clear behavioral transitions as a proxy for information, we can explicitly track the spread of this “information” through the group.  Simplifying the problem in this way allows for a direct comparison of models for the interaction network over which this information propagates. 

In the experiment, the informed fish were trained to associate a green laser with food, such that when the light turned on they would swim rapidly towards it. Some of these informed fish were then placed within a group of uninformed (untrained) fish.

The video below shows one example of what happened when the light was switched on. Informed fish are indicated by red dots and uninformed by blue dots. During a "leadership event," such as the one shown below, each fish exhibits a characteristic "response" in which it begins to move quickly toward the target. These responses are indicated in the video below as red circles that appear around each fish at the time of its response, and they can be seen to propagate through the group in a clear wave of information transfer. 

Using these data, we can test different models for what predicts when a fish will respond. In particular, we are interested in uncovering the (time-varying) interaction networks over which information spreads in the school. In other words, who interacts with whom in the group? 

For more information, see our related web feature or read the paper.