It’s not uncommon to see a flock of birds, startled by some perceived threat, take to the air in a highly coordinated flight. Similar behavior can be observed in schooling fishes where each fish, mimicking the movement of its neighbor, turns, darts or zigzags away from a threat with uncanny precision. In both cases, it seems as if the respective groups — the flock and the school — are acting as a single, unified entity.
This phenomenon, called a “behavioral cascade,” has long confounded biologists. How decisions about which way to move and how fast are communicated to many individuals — often numbering in the hundreds or even thousands — and how these decisions are processed and coordinated so swiftly is a little understood topic.
In a paper published on Sept. 23 in Proceedings of the National Academy of Sciences, Matt Sosna, who received his Ph.D. in ecology and evolutional biology from Princeton and is a data scientist at Aquicore, and his colleagues investigate the central question of how group animals — like flocking birds, schooling fish and herding mammals — process information about their environment, a procedure called collective computation. To do this, the researchers studied how small schooling fish called golden shiners (Notemigonus crysoleucas) respond to perceived threats. Like flocking birds and herding ungulates, these fish are known for evading predators by exhibiting “startle cascades.”
This behavior is analogous to an avalanche or a wave, said Sosna. It begins by a fast and sudden burst of swimming initiated by one or more fish responding to a perceived threat. “Think about a soccer game where everyone is doing the ‘wave,’” he said. “I stand up and then you stand up. And then the next person stands up. You can think of this as a behavioral cascade so that the behavior — standing up — is cascading through the group like a wave.”
Shiners, common to ponds, lakes and sloughs in eastern North America, are a “jumpy” species that startle often — sometimes even apparently randomly — when there’s no overt threat. Sometimes two or three or more fish will startle, but that does not necessarily propagate throughout the school.
However, a large cascade — one that spreads through the whole school — depends on how close the fish are together. Shiners will bunch up closely in response to a threat. This in itself does not cause a cascade but, because the fish are packed closely together, when a startle does occur, Sosna noted, it is more likely to generate a full-blown startle cascade.
“It’s kind of like if you’re across a movie theater from me and you yell and startle. I might not startle, but if you’re right next to me I will,” said Joseph Bak-Coleman, a graduate student in ecology and evolutionary biology, who also contributed to the paper.
The necessity of avoiding predators and other threats is a crucial aspect of survival. It is a key element, along with mating and food acquisition, dictating an organism’s reproductive success or “fitness” — its ability to pass on its genes to succeeding generations. Decision-making among individual animals faced with risky or potentially dangerous situations — such as from a predator or some other threat — has been well studied. Scientists know quite a bit about how the brains of individual animals respond to such contingencies. They do this, for example, through a complex interplay involving sensory inputs from the environment and previous experience — all mediated by physiological states and limitations.
The important point is that the individual animal is responding directly to its environment. It is modifying its behavior based on the particular context in which it finds itself. For example, a grazing gazelle on the African savannah will take to flight if soon after it hears a lion roar it notices the bushes nearby rustling. If, however, rustling bushes are not preceded by a lion’s roar, the gazelle will likely continue to graze rather than take off running.
But collective or group behavior under risky conditions, especially group decision-making that occurs with often split-second timing, is a phenomenon of a different order — and one that has been little studied.
“We often tend to think intelligence is in our brains — that it is in the individual organism. We don’t often think that intelligence can be encoded by changing the strength or structure of social interactions between organisms,” said Iain Couzin, director of the Max Planck Institute of Animal Behavior in Konstanz, Germany, a visiting research scholar in ecology and evolutionary biology, and the senior author of the study.
A group response is often reflected in the ability of the group to incorporate all these individual responses and, in some cases, override or transcend them.
“It’s as if the fish are acting like a collective brain,” Sosna added, “so that the structure of the network can change extremely quickly. They’re changing the network in a way that allows them to be responsive to the environment.”
It’s not that individuals necessarily change their behavior. They will still react by assessing the threat and acting accordingly — either through reacting or doing nothing. But the important aspect is the way all those individual behaviors get translated into collective behavior.
To investigate just how this translation might occur, Sosna and his fellow researchers conducted controlled experiments using shiners in a large tank. The team set up cameras above the tanks to record the action and movement of the fish. They added a chemical called Schreckstoff, literally “scary stuff” in German. This substance is derived from the skin cells of fish and commonly used to elicit a fear response.
The cameras recorded the responses of the fish down to a resolution of 120 frames per second. These data were then analyzed using computer vision software, which tracked and recorded the trajectory of each fish and estimated its body posture as well as its visual field.
The researchers found that the application of Schreckstoff to the tank resulted in the fishes bunching up closely together and an increase in the incidents of startle cascades. Most importantly, however, the researchers also found that Schreckstoff did not increase the likelihood of a fish responding to its neighbor’s startle.
This means that an individual fish does not change its behavior; the way each fish responds to its neighbor remains the same. What changes is the degree of the relationship between the fishes. In other words, it results from the structure of the group itself rather than how any individual fish in the group or social network perceives and responds to its environment.
“By changing the structure of the group, by coming closer together, they make the strength of the social connectivity among these individuals much stronger,” Couzin said.
“An individual in the group doesn’t necessarily need to know what’s going on,” Sosna added, “but as long as the group itself is doing the computation, then the individual gets the benefit of being in that group and avoiding predators.”
This strategy — moving closer to heighten sensitivity to threat and moving farther apart to lessen sensitivity — may have evolved, Sosna said, because it is much easier than the more complex mental computations required by individuals facing threats.
“You can see all these really interesting group-level patterns emerging,” Sosna said, “but all this still has to boil down to the benefit to the individual. The fish are all completely, selfishly, trying to survive.”
The researchers believe that the results from the experiment could lead to new insights into the collective behavior of animals as well as have wider practical applications. Biological group systems, like the fish school studied here, have been selected over millions of years to process information, Couzin said, “so we can really learn something fundamental about the relationship between the structure of these networks and how to effectively process information.”
The insights gleaned here might help engineers in the development of technologies that can be applied to problems demanding more effective and efficient solutions. For example, such insights might help designers in creating a system where, collectively, a swarm of robots could manifest a “collective intelligence,” even when each individual robot is programmed to carry out simple tasks.
“If we can network [these robots] together in a similar way animals or groups network, we can achieve a collective intelligence, an ability for them to respond appropriately in a changing world,” Couzin said.
The paper, “Individual and collective encoding of risk in animal groups,” by Matthew M.G. Sosna, Colin R. Twomey, Joseph Bak-Coleman, Winnie Poel, Bryan C. Daniels, Pawel Romanczuk and Iain D. Couzin, was published on Sept. 23 in Proceedings in the National Academy of Sciences (DOI: 10.1073/pnas). This work was supported by an NSF Graduate Research Fellowship, a MindCORE Postdoctoral Research Fellowship, and the Deutsche Forschungsgemeinschaft (DGF, German Research Foundation).