This link will take you to the Google Doc used to create the chapter. Below is a copy lifted from the link on March 3, 2025.
Collective behavior in animals: an ecological perspective
Deborah M. Gordon, Dept of Biology, Stanford University
Systems without central control are ubiquitous in nature. Cells act collectively – as networks of neurons to produce thoughts and sensations, or as T-cells and B-cells that together mobilize immune response to pathogens. Many animals act collectively, as bird flocks that turn in the sky, or fish schools that swerve to avoid predators. Social insects live in colonies, and groups work together to collect food, build nests and care for the young.
Nature offers a huge variety of examples of collective behavior in animals (reviewed in Sumpter 2010). Because hierarchy is so familiar in human organizations, it may at first seem surprising that animal groups can function without any central control. Much of the study of collective behavior in animals has been devoted to establishing that there really is no leader, and to considering how this could possibly be true. For example, in early experiments on the formation of a “V’ by a flock of geese, the investigator shot the goose at the front of the V. Another goose immediately took its place, demonstrating that the V shape was not the result of leadership on the part of a particular animal, but instead arises from interactions among all the geese regardless of each goose’s personal qualities.
Modelling animal collective behavior
Most work so far on collective behavior in animals seeks to describe how groups of a particular species accomplish a certain task: how fish schools turn, or how ants recruit others to food. The answer to this question is an algorithm that explains, in general, how local interactions produce a particular outcome.
Recent work has made significant progress in identifying the algorithms used in animal collective behavior (Sumpter 2010). Over the past 20 years, across all the fields of biology, attention has turned to deciphering how local interactions create global outcomes. The goal of the first models of collective behavior was to demonstrate that such behavior, called ‘emergence’ or ‘self-organization’, can exist, by showing how it works in a particular case (e.g. Deneubourg et al. 1986). A growing field of research continues this effort, seeking to describe the local interactions that give rise to some collective outcome (e.g. Couzin 2005). Patterns of animal movement are the best studied form of collective behavior in animals. Models of the movement of fish schools explain how local interactions among fish allow the school to turn, without any decision from any individual about where to go (Handegard et al 2012 ); studies of the movement of locust groups explain how interactions between individuals push the group from the stationary to the migratory phase (Bazazi 2012).
So far the study of collective behavior in animals has been devoted to discovering what are the algorithms that allow local interactions to produce collective outcomes. Enough progress on this has been made that it now may be possible to sort types of collective behavior, according to the algorithms that generate the behavior.
One set of models, mainly applied to social insects, describe collective behavior as ‘self-organization’ (Nicolis and Prigogine 1977). These grew from a search for analogies in animals to transitions shown in thermodynamics by Prigogine, in which accumulated interactions lead to an irreversible outcome. All that is needed is some local interaction in which one individual influences the behavior of another, such as following, repelling, or imitating. Pour some water on the counter and small streams will lead from the main source and find a way to drip off the edge. The direction that the streams will take is the result of local interactions among molecules: when enough go in the same direction to overcome the surface tension of the water, a stream will form. In the same way, ants that use trail pheromone will form trails to resources. From a central nest surrounded by scattered resources, the direction that trails will form depends on local interactions among ants. To get started, there must be some random movement by searching ants. If some ants put down a signal, such as a chemical trail, when they find food, then eventually trails will form in the directions that happened to attract enough ants to pull in other ants that get to the food and reinforce the trail in time to attrach others.
Another class of algorithms comes from networks in which the rate of interaction, rather than any particular message conveyed in an interaction, is the cue that provides information in a network. For example, harvester ants use simple positive feedback, based on the rate of interaction between outgoing foragers and foragers returning with food, to regulate foraging activity (Gordon et al. 2011, Pinter Wollman et al. 2013).
Does all collective behavior in animals qualify as collective intelligence? Sometimes we call collective behavior ‘intelligent’ when it seems to serve a purpose. Watching ants, it’s clear that ants can be quite stupid, yet colonies manage to adjust to appropriately to subtle changes in conditions, and ants have been extremely successful worldwide many millions of years. Sometimes we call collective behavior ‘intelligent’ when it seems to be an example of behavior that we already think is intelligent, such as memory or learning. This follows Turing’s lead; we can agree that anything is a brain if it performs particular functions. But to avoid tautology, it is important to keep in mind that the resemblance to intelligence is established by definition. For example, we could ask whether ant colonies learn or remember, and the answer will depend mostly on whether we define learning or remembering to include some things that ant colonies do. In the following I will use ‘collective behavior’ interchangeably with ‘collective intelligence’.
We do not yet have a general theory of collective behavior in animals. The growing body of work that shows how local interactions produce collective behavior feeds the hope that there may be general laws that could explain how groups work together. Similar algorithms have been used to describe collective behavior in disparate systems. Systems biologists have noticed that some network configurations, or motifs, occur more often than others, and that these may be due to basic principles that make certain configurations more efficient (Doyle and Csete 2011, Alon 2007). Agent-based models draw on the results in statistical mechanics that predict the behavior of a group of particles from the Brownian motion of a single particle (Schweitzer 2003). Such models provide ways to describe many kinds of collective behavior, from molecules to the cellular processes that create patterns in the course of development, to the movements of ants or migrating wildebeest. Of course, demonstrating that two different processes can be described by the same model does not demonstrate that the processes are alike. Any process can be described by many different models.
Ecology of collective behavior
To learn whether there are general laws of collective behavior that apply across different species and systems, we will need to ask new questions about the ecology of collective behavior and how it evolves. The population geneticist T. Dobzhansky once pointed out that nothing in biology makes sense except in the light of evolution. A corollary is that nothing in evolution makes sense except in the light of ecology. Evolution occurs as a result of the relation between a phenotype and its environment. The next step in the study of collective behavior in animals is to consider the ecology of collective behavior and how it evolves.
Collective behavior is the result of evolutionary processes that shape behavior to modify and respond to environmental conditions (Gordon 2014). Investigating how these algorithms evolve can show how diverse forms of collective behavior arise from their function in diverse environments. Wildebeest migrate to get into an environment with grass to eat; bees swarm when one nest is too small for the hive, making the old one large enough for the bees that stay behind and putting the new swarm in an appropriate place. Fish schools turn, and bird flocks move, in response to predators; schools of dolphins chase their prey, even working together to herd fish into fishermen’s nets in return for food (Pryor and Lindbergh 1990 ).
We can consider the evolution of collective behavior in the same way we think about the evolution of other traits. Rather than asking what do individuals have to sacrifice to participate in group activities, we can ask how does the collective behavior influence the survival and reproduction of the participants and of the entire system.
To learn about the evolution of collective behavior, we must begin to consider variation, because this is the starting point for natural selection. To consider the evolution of individual behavior, we look for differences among individuals. In the same way, to consider the evolution of collective behavior, we must look for variation among collectives. Although we speak of the algorithm that a school of fish or an ant colony or a human group uses to accomplish some task collectively, in fact every school, colony or group behaves differently. Natural selection can occur when this variation is inherited and has some impact on the reproduction of more fish, ants or people.
Ant colonies are an ideal system in which to consider the evolution of collective behavior (Gordon 2010) because colonies are the reproductive individuals. A colony consists of one or more female reproductives, or queens, who produce the workers, all sterile females, and also sons and daughter queens. Colonies reproduce when reproductive males and daughter queens from different colonies mate, and the newly mated queens go on to found new colonies. The ants in the colony work together to produce reproductives, who mate with the reproductives of other colonies and found new colonies. Thus colonies produce offspring colonies, as a result of their collective behavior. Natural selection can shape the way that individual ants work together in colonies to produce more colonies. This evolutionary process starts with variation among colonies.
Collective behavior is sometimes equated with cooperative behavior, but these are not the same thing. To identify behavior as collective is to focus on the outcomes at the level of a group of participants. There are many examples of collective behavior in which the participants are not sentient individuals, such as brain functions produced by the collective behavior of neurons. There are other examples of collective behavior in which the outcomes occur regardless of the motivations and intentions of the participants; a familiar example is a traffic jam. So animals that engage in collective behavior are not necessarily acting according to an intention or motivation to work together.
Cooperation is often defined in terms of costs and benefits: the cooperating individual sacrifices to be part of the group. This compares participation in a group with an imaginary prior state in which the individual functions on its own. An example of this view comes from early work on social insects. In a social insect colony, many individuals do not reproduce, but particpate collectively in activities that allow other individuals to reproduce. A colony reproduces by producing reproductives that found new colonies. It was suggested that this system could not evolve unless all the individuals shared whatever genes might be associated with worker sterility; otherwise such genes could not persist. Since then genetic studies have shown that relatedness among social insect workers is not as high as originally supposed, indicating that however the system originated, it is not currently maintained through kin selection. More important, it is misleading to frame worker sterility as a sacrifice by individuals (Gordon 2012a). Reproduction as a colony entails a sacrifice on the part of individual social insect workers only in the sense that reproduction as a human entails a sacrifice on the part of somatic cells such as those in liver and bone.
Ant colonies vary in behavior. For example, harvester ant colonies differ in the regulation of behavior (Gordon et al 2011) . Some colonies year after year are less likely to forage in poor conditions, when low humidity means a high cost in water loss, and when little food is available. This arises from differences among colonies in the collective behavior that regulates foraging based on between returning foragers, bringing food into the nest, and outgoing foragers.
Harvester ants foraging in the desert lose water when walking around in the hot sun, and they get their water only from the fats in the seeds they eat. Thus the ants must spend water to get water. Faced with the high operating costs of the network that regulates foraging, evolution has opted for positive feedback, so that the system keeps running unless something stops it. Harvester ants use the rate at which ants are returning with food as the source of positive feedback. Each ant returns from its foraging trip, puts down its seed and waits inside the nest. It does not go out until it meets returning foragers with seeds at a high enough rate. The more food is available, the more rapidly ants will come in with seeds. When food is scarce, ants take a long time to find food and trickle back to the nest. Few of the foragers in the nest then go out, so there is little water loss.
Transmission Control Protocol (TCP), which manages traffic congestion in the internet, uses a similar algorithm based on positive feedback (Prabhakar et al 2012). Each time a data packet succeeds in arriving at a node, an acknowledgement is sent back to the source, and this stimulates the transmission of the next data packet. The cost of sending data when there is no bandwidth is high, because the data might be lost. Using this system, data transmission regulates itself to match the available bandwidth.
Variation among ant colonies in collective behavior is due to variation among colonies in how their individual ants respond to interactions (Pinter Wollman et al 2013). We are currently using models from neuroscience to describe exactly how ants differ in the ways that they assess their rate of interaction, by analogy to the way that neurons act as ‘leaky integrators’ to asses their rate of electrical stimulation (Goldman et al. 2009).
Variation among colonies in the regulation of foraging appears to be inherited from mother to daughter queens. A colony lives for 20-30 years, as long as its founding queen survives. All the ants in the colonies are the offspring of the founding queen. However, the worker ants live only a year. Year to year trends in the collective behavior of the colony thus appear to be inherited in successive cohorts of workers from their mother, the colony’s queen. Moreover, it seems that parent and offspring colonies are similar in their sensitivity to current conditions, and in their collective decisions about when conditions are poor enough to deter the ants from foraging (Gordon 2013).
Recent work on harvester ants shows that natural selection is shaping collective behavior so as to conserve water (Gordon 2013). The harvester ant study site in Arizona is subject to the severe and deepening drought throughout the southwestern US in the past 10-15 years. Colonies that are less likely to forage in poor conditions, when water loss is high and the return in food is low, tend to have more offspring colonies. Variation among colonies in the collective regulation of foraging, the heritability of this collective behavior, and the impact on colony reproductive success, in offspring colonies, all are shaping the evolution of the collective intelligence of the colony.
Ants are an extremely diverse group of more than 12,000 species that occupy every conceivable habitat on Earth. The diversity of environmental constraints has led to diverse tactics in the collective regulation of colony activity (Gordon 2014). For example, while some species search for and retrieve food from a central nest, others circulate along highway systems of permanent trails and search for and retrieve food from the trails (Flanagan et al 2013, Gordon 2012b).
Similar environmental constraints may lead to the evolution of similar forms of collective behavior in different systems (Gordon 2014). Collective behavior allows animal groups to respond to environmental challenges such as operating costs, the distribution of resources, and the stability of the environment. To examine how collective behavior evolves, we can ask whether we see patterns in the response to particular environmental constraints.
Empirical studies in many species are needed to learn how environmental conditions are shaping the evolution of collective behavior. A quantitative description of why a process is effective, or a simulation that selects for that process, helps us to understand how it works. But to understand its evolution we need to know its ecological consequences, what problems it solves in a particular environment, and how it is shaped by, and influences, changing conditions.
Outlining hypotheses about the fit between collective behavior and its environment can guide the investigation of collective behavior. For example, we now know enough about physiology that we expect animals that live in hot places to have adaptations for heat exchange. In the same way, we can expect the algorithm that dictates collective organization in particular conditions to be tuned to the constraints of those conditions. With respect to the workings of collective biological systems, we are like the European naturalists of the early 19th century, agog in the Amazon. We are searching for general trends amidst enormous diversity and complexity. A framework for the match between process and environmental conditions can provide predictions that guide the investigation of new systems.
To study the evolution of collective behavior, we need to consider variation among groups in how the algorithm that regulates behavior is deployed, and whether such variation is associated with differences in the ecological success of the collective. Evolution does not always produce the optimal state, and often what is optimal is not apparent. Genetic algorithms can show how evolution could happen, but the selective pressure is created by the programmer, not by nature.
We’d like to know whether there are general principles of collective behavior that apply across systems. A first step is to classify the algorithms used, but this is only the beginning. The next step is to consider how collective behavior evolves in natural systems (Gordon 2014). This may reveal trends in the kinds of algorithms used in particular environments. To study the evolution of collective behavior empirically, we need to employ the same methods used to study the evolutionary ecology of individual traits. How do groups vary in their use of a collective algorithm? How does this variation matter to the success of the group? What’s needed are investigations of how collective behavior functions in its environment.
References
Alon U (2007) An introduction to systems biology: design principles of biological circuits. Boca Raton (Florida): Chapman & Hall/CRC. 301 p.
Bazazi, S., Bartumeus, F., Hale, J.J., Holmin, A.J. & Couzin, I.D. (2012) Intermittent motion in desert locusts: behavioral complexity in simple environments. PLoS Computational Biology, 8(5), e1002498.
Couzin ID, Krause J, Franks NR, Levin SA (2005) Effective leadership and decision-making in animal groups on the move. Nature 433: 513–516.
Deneubourg JL, Aron S, Goss S, Pasteels JM, Duerinck G. (1986). Random behavior, amplification processes, and number of participants: how they contribute to the foraging properties of ants. Physica D 22:176-86.
Doyle JC, Csete M (2011) Architecture, constraints, and behavior. Proc Natl Acad Sci U S A 108 Suppl 3: 15624–15630.
Flanagan TP, N M Pinter-Wollman, M E Moses, D M Gordon. (2013) Fast and flexible: Argentine ants recruit from nearby trails. PLoS One: 8(8): e70888. doi:10.1371/journal.pone.0070888.
Goldman MS, Compte A, Wang XJ. (2009). Neural integrator models. Encycl Neurosci.6:165-178.
Gordon DM (2010) Ant Encounters: Interaction Networks and Colony Behavior. Princeton (New Jersey): Princeton University Press. 167 p.
Gordon DM, Guetz A, Greene MJ, Holmes S (2011) Colony variation in the collective regulation of foraging by harvester ants. Behav Ecol 22: 429–435.
Gordon, D. M. (2012a). What we don't know about the evolution of cooperation in animals. In Cooperation and Its Evolution, Kim Sterelny, Richard Joyce, Brett Calcott, and Ben Fraser, Editors. Cambridge, Massuchesetts, The MIT Press.
Gordon DM (2012b) The dynamics of foraging trails in the tropical arboreal ant Cephalotes goniodontus. PLoS ONE 7: e50472.
Gordon, D.M. (2013). The rewards of restraint in the collective regulation of foraging by harvester ant colonies. Nature. DOI: 10.1038/nature12137
Gordon DM. (2014). The ecology of collective behavior. PLoS Biol 12(3): e1001805. doi:10.1371/journal.pbio.1001805
Handegard, N.O., Leblanc, S., Boswell, K., Tjostheim, D. & Couzin, I.D. (2012) The dynamics of coordinated group hunting and collective information-transfer among schooling prey. Current Biology, 22(13), 1213-1217
Nicolis G and Prigogine I. (1977) Self-organization in Non-equilibrium systems. John Wiley and Sons, NY.
Pinter-Wollman, N., Bala A., Merrell A., Queirolo J., Stumpe M.C., Holmes S., D. M. Gordon. (2013) Harvester ants use interactions to regulate forager activation and availability. Animal Behaviour 86:
Prabhakar, B., Dektar, K.N., and D.M. Gordon. (2012). The regulation of ant colony foraging activity without spatial information. PLoS Computational Biology 8(8):e1002670. DOI:10.1371/journal.pcbi.1002670
Pryor, K. and Lindbergh, J. (1990). A dolphin-human fishing cooperative in Brazil. Marine Mammal Science, 6: 77–82. doi: 10.1111/j.1748-7692.1990.tb00228.x
Schweitzer F (2003) Brownian agents and active particles: collective dynamics in the natural and social sciences. Berlin; London: Springer. 420
Sumpter DJT (2010) Collective animal behavior. Princeton (New Jersey); Oxford: Princeton University Press. 302 p.
Comments