Review Collective cognition in animal groups Iain D. Couzin Department of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ 08544, USA The remarkable collective action of organisms such as with mechanisms of decision-making within the brain swarming ants, schooling fish and flocking birds has [1,6]. Although many details differ, there is good reason long captivated the attention of artists, naturalists, phi- for increased communication between researchers inter- losophers and scientists. Despite a long history of scien- ested in collective animal behavior and those in cognitive tific investigation, only now are we beginning to science. decipher the relationship between individuals and group-level properties. This interdisciplinary effort is Collective motion beginning to reveal the underlying principles of collec- It is usually not possible to scale reliably from individual to tive decision-making in animal groups, demonstrating group behavior through verbal argument alone. Con- how social interactions, individual state, environmental sequently, considerable progress in revealing the prin- modification and processes of informational amplifica- ciples of collective behavior has been made using tion and decay can all play a part in tuning adaptive mathematical modeling techniques, such as computer response. It is proposed that important commonalities simulation (Box 1). Some of the earliest theoretical exist with the understanding of neuronal processes and approaches were inspired by particle physics [2,7]. These that much could be learned by considering collective introduced the influential concept of using equations to animal behavior in the framework of cognitive science. characterize individual movements and interactions (as ‘social forces’), the aim being to explore whether (and if so, Introduction how) individual behaviors can scale to the coherent collec- It is little wonder that the behavior of animal groups, tive motion exhibited by fish schools or bird flocks. such as schools of fish, flocks of birds or swarms of insects Valuable insight was gained using such model descrip- has been associated with the concept of having a ‘collec- tions. First, it was realized that collective behavior can tive mind’ [1]. Grouping individuals often have to make arise from repeated and local interactions and need not be rapiddecisionsaboutwheretomoveorwhatbehaviorto explicitly coded as a global blueprint or template [2,8–10]. perform, in uncertain and dangerous environments. In addition, biologically plausible local interactions can Decision-making by individuals within such aggregates account for the typical group structures found in nature is so synchronized and intimately coordinated that it has (Box 1). Because of the nature of these local interactions, previously been considered to require telepathic com- behavioral control is typically distributed, as opposed to munication among group members or the synchronized control being hierarchical with one (or a few) leader(s) response to commands given, somehow, by a leader controlling group-members’ actions. [2,3]. Because distributed coordination does not depend on a In fact, individuals base their movement decisions on specific subset of individuals, groups are inherently robust locally acquired cues such as the positions, motion, or to perturbation. Analogous decentralized principles govern change in motion, of others [2], making the collective the coordination of many neuronal assemblies, enabling response all the more remarkable. Each organism typically robust encoding of information across a wide range of has only relatively local sensing ability (further limited in spatial and temporal scales [11–22]. Information from large aggregates by crowding). Groups are, therefore, often multiple distributed sources can be acquired and processed composed of individuals that differ with respect to their simultaneously, thus allowing individual (cells or organ- informational status and individuals are usually not aware isms) access to computational capabilities not possible in of the informational state of others, such as whether they isolation. are knowledgeable about a pertinent resource, or of a A further principle revealed by computational modeling threat [1,2,4,5]. of grouping, as outlined in Box 1, is that multiple stable Recent studies have begun to elucidate how the modes of collective behavior can co-exist for exactly the repeated interactions among grouping animals scale to same individual interactions [10]. This is directly analo- collective behavior, and have revealed, remarkably, that gous to multistability in neural systems, in which multiple collective decision-making mechanisms across a wide collective states (attractors) co-exist at the same value of range of animal group types, from insects to birds (and the system’s parameters both within neurons themselves even among humans in certain circumstances) seem to and in neural networks [20]. Multistability in neural sys- share similar functional characteristics [2,4,5]. Further- tems has been suggested as an important mechanism for more, at a certain level of description, collective decision- memory storage and temporal pattern recognition [20], making by organisms shares essential common features and an intriguing (but currently untested) possibility is that similar functional benefits might exist for animal Corresponding author: Couzin, I.D. ([email protected]). groups. 36 1364-6613/$ – see front matter ß 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2008.10.002 Available online 6 December 2008 Review Trends in Cognitive Sciences Vol.13 No.1 Amplification and damping of collective response made can often be the result of amplification of random Feedback processes fluctuations or stochastic initial conditions. This can result The social context created by highly integrated behavior in ‘informational cascades’ in which arbitrary choices are strongly affects the way information is acquired, transmitted made [2,5,27]. Incorporating negative feedback can prevent and processed by group members. Specifically, it can facili- such over-sensitivity of collective response to individual tate the collective amplification and damping of information error or environmental noise and can enable long-range and, thus, the adaptive tuning of collective behavior in patterns to be detected in the face of distracting local response to external stimuli and/or internal state. fluctuations [28]. Typically, however, this ability comes at Alignment among individuals (a tendency to move in the a cost as the time taken to make a decision is increased (a same direction as near-neighbors, Box 1), for example, can speed–accuracy trade-off). enable information about a change in direction to be Little is currently known about how vertebrate groups, transmitted as a rapid wave of turning, over long distances such as fish, birds or herding quadrupeds actually balance [23,24]. Amplifying local fluctuations through positive the trade-off between speed and accuracy during collective feedback is important when threats, such as predators, decision-making. A clue comes from an experimental study are detected because it creates an ‘effective’ sensory range by Ward et al. [25] on schooling stickleback fish (Gasteros- much greater than that of individual perception [1,2]. For teus aculeatus). Individuals were shown to exhibit a highly example, the change of direction of only a few individuals non-linear response to near-neighbors; they largely disre- that initially detect a cryptic predator can be amplified gard the movement decisions of a single neighbor, but rapidly, as a propagating wave of turning, resulting in a strongly increase their probability of copying as more neigh- much larger number of individuals, or even a whole group, bors (a ‘quorum’) commit to a given direction of travel. This turning away from the threat [1,2,23,24]. functional response was found to improve the accuracy of Similar amplification processes are a fundamental com- individual decision-making by enabling fish to integrate ponent of neural information-processing. They facilitate their own estimation with that of others, while not incurring the translation of local stimuli to response both within the much cost in terms of the time taken to make a movement cell (the conversion of localized ion fluxes to action poten- decision. tials) and to intensification of propagating electrical and This form of non-linear response represents a common chemical activity across networks of cells, such as the theme among many decision-making systems that face traveling waves of activity seen in the vertebrate cortex speed–accuracy tradeoffs (see later discussions on social [13,17,20–22]. insects). An initial slow phase enables appropriate infor- A further commonality with neural signal propagation mation to be accumulated before a transition to a higher and information encoding is that it is difficult for animals in commitment to one option, among alternatives. groups to tune collective response with positive feedback A similar challenge faces the brain when presented with alone. Damping, or negative feedback, is often an important ambiguous conflicting sensory stimuli. Within the primate regulator of group response. Tuning, adaptively, collective brain, for example, sensory evidence for different stimuli is response through regulation of the relative influence of encoded as firing rate within separate, competing neural positive and negative feedback is the essence of decision- groups [11,12,14–16,18,19].