Stephen C. Pratt, Eamonn B. Mallon, David J. T. Sumpter, Nigel R. Franks ... Nathalie Stroeymeyt, Elva J.H. Robinson, Patrick M. Hogan, James A.R.. Marshall ...
Collective Intelligence: How does a group outdo its members?
Collective intelligence is a group of agents acting together as a single cognitive unit1. It is characterized as coordinated but with no central control, without a leader, and is an example of self-organization—collective decisions emerge from interactions between many individuals acting locally based on simple rules, without global knowledge2. While collective intelligence has been extensively studied in social insects, it is also found in numerous other systems, from the cellular level, to complex vertebrate societies. This paper outlines the evidence suggesting that collective decisions are more accurate than individual decisions, with the goal of describing the mechanisms and pitfalls of collective intelligence, as well some applications of its study. When making a decision, individuals balance both personal information, such as past experience, and social information, such as cues from neighbors3, allowing an accurate consensus to be reached. Individual decisions are influenced both by a drive to protect oneself, but also by a drive towards group cohesion, which can integrate multiple dimensions of a problem, allowing group decisions to be made accurately about choices that individuals are not even explicitly aware of. This social awareness, and the ability to indirectly detect neighbor’s preferences, achieves accuracy that is close to provably optimal4. “Optimality” however, is a relative term, and deserves discussion. Keeping a group together, or maintaining ‘group cohesion’, has been used as a measure of a decision’s ‘precision’ – when the rock ant Temnothorax albipennis is faced with different time constraints, a decision can be made quickly but non-cohesively, or slowly and cohesively5. This speed-precision trade-off has been observed in cooperative hunting, collective navigation and escape behaviors. It is especially useful for adaptive responses to drastically different situations, such as careful selection of a new nest when time is plenty, and fast abandonment of a current nest in a crisis6, or when honeybees need to optimally time the construction of their new comb in a way that ensures there is enough space of nectar when it arrives without expending too much energy in construction too early7. It is important when discussing collective intelligence and its advantages to define how exactly the success of a decision is defined, and the literature contains various suggestions that do not always overlap, so a robust definition of precision still needs work. To understand the mechanisms of collective decision-making, it is useful to look at what some have claimed to be one of the most difficult and complex collective choices—namely, house hunting by social insects8. Temnothorax curvispinosus colonies compare options, and are able to both reject a mediocre site in favor of a good one, and choose that same mediocre site rather than a worse one, so they can pick the best option9, and their behavior qualitatively matches that of Temnothorax albipennis. This is achieved in the following way:
when a scout finds a potential site, there is a delay inversely related to that site’s quality, followed by initially slow tandem runs, where the scout recruits other active ants individually to follow it to the found site, and once a certain threshold of ants arrives at that site, a fast transport of the passive majority begins, where other ants can be picked up and moved directly to the new site. Thus, three of the key characteristics that increase a decision’s chances of being optimal are an initial delay, a slow start, and a fast acceleration. The delay, which is also found in honeybee house hunting, means more time is left for the potential discovery and comparison of alternatives8. The fast nonlinear acceleration of migration, where transport occurs three times faster than tandem runs, allows a rapid decision to be made once enough ants have been recruited10. Even the slow linear recruitment during tandem runs is important, however. During this phase, signal effectiveness is proportional to signal quantity, and yet when isolated, this alone is enough for colonies to successfully choose between two unequal feeders, and even reallocate recruitments when those feeders are switched11. Therefore, successful collective decisions can outdo individual decisions when there is enough time for comparisons, there is a steady baseline flow of information, and action is taken quickly once the criteria have been met. A reliable exchange of information between members of a group is particularly important for a successful collective decision. Marine predators coordinating hunting groups rely both on their own collective behavior, say, in maintaining some attack formation, and on their ability to disrupt their prey’s collective behavior by fragmenting the prey group structure and inhibiting their collective information transfer12. Therefore, collective intelligence is a fundamental part of both predator and prey strategies, and it relies on easy information transfer. Furthermore, interactions between individuals and rules of evaluation need to be simple for this information transfer to be successful. Individual ants are not likely to personally compare different nests, but the simple threshold rule described above is sufficient for rational group decisions, and it only involves the simple interaction of recruitment13. Similarly, during collective evasion maneuvers of schooling fish (Notemigonus crysoleucas), individuals rely on simple, robust rules to assess their neighbor’s changing behavior, and that is enough for complex group behaviors to arise14. Clearly, successful collective decisions hinge on efficient and simple communication between members of a group. Before moving on to other conditions of collective choice, it is worth discussing another important instance of such decisions in social insects— foraging behavior. Ant colonies need to modulate foraging effort with changing demand, and this flexibility is maintained by distributing a range of response thresholds to foraging stimuli, such as brood hunger signaling, or number of returning ants15. This means that most of the time, when the threshold is low, only a few foragers will bring in food, but when demand increases, more foragers will have their threshold reached and more food will be brought in. It has been
shown that, on the individual level, completing a task lowers the threshold for that task, so that individual gains experience and is more likely to do that task again. Thresholds depend on “corpulence”, so the less fat reserves an ant has, the lower its threshold, and the more likely it is to go forage. This, in its significance to the collective decision to increase a group’s foraging workforce, is parallel to the simple local evaluation and recruitment carried out by scouts and that behavior’s significance in the collective decision to select a new nest and emigrate to it. Corpulence is actually social information too, since it relays information about the general state of the colony and its recent access to food. Clearly, many of the key choices made by social insects take the form of collective decisions. The role of well-informed individuals in collective decisions has long been a topic of debate. It is a particularly apt question, when often the perceived selling point of collective intelligence as a concept is that the group is seem as smarter than its smartest member. Small societies, such as those of Temnothorax albipennis, have been found to somewhat rely on well-informed individuals16. While these are not “leaders” per se, they are individuals who have, for example, visited both potential nest sites, and can actively choose the better one. While the majority of the ants do not actually encounter both sites, the colony does takes advantage of this direct decision by its well-informed members. In the case of schooling fish, certain key individuals in frontal positions are essential to successful group decisions, and how well they balance personal information and social cues drives that success. Somewhat counter-intuitively, it has also been found that the more uninformed individuals in a group, the better the collective decision turns out to be. When there are two small groups of informed individuals who disagree about which direction the group should move towards, adding more uninformed members actually improves the stability of the decision, meaning most of the group chooses one of the two options17. When the decision is unstable, there is a loss of group cohesion, which as suggested before, is often used as a measure of success for a given collective choice. Therefore, increasing the number of uninformed individuals increases the stability and the success of the decision. Even in more complex, socially stratified societies, collective action is present. When their group needs to move, wild baboons prefer following a group of initiators rather than a dominant individual, and analogously to group movements described above, they will choose one direction over another if the angle between the two options is above a certain threshold, but will compromise if it is not18. Humans also exhibit collective behavior, both in small and large groups19. Decision speed and likelihood of group splitting has bee found to depend on group size and presence of uninformed individuals, with accuracy hinging on the spatial positioning of informed individuals. As species’ complexity increases, any collective decisions will clearly become more complicated, but there is also evidence that some of the same factors that fuel collective decisions social insects and marine animals are present in monkeys and humans.
The perceived prevalence and success of collective intelligence has birthed the term “wisdom of the crowd”. Tom Seeley, in his signature book Honeybee Democracy, outlines the following four conditions that make a crowd “wise”: independence, diversity of opinion, and decentralization20. These fit well into the characteristics of success described so far. Diversity, which allows a wide range of possible solutions to be proposed, is promoted by the delay in recruitment seen in ants and honeybees. Decentralization is seen in the lack of a leader, and in the opinion polling seen during house hunting by social insects. Independence, with the goal of preventing “information cascades” where a bad decision is followed by increasingly many members of a group, such as ants separated from a colony following each other in a circle until they die of exhaustion21. Having many decisions be made at the same time, rather than all at once, is one way to increase independence, and this is encouraged by simple, robust local rules for behaviors, such scouting in social insects. In many cases, these conditions are met. However, collective decisions are not always rational, and mistakes can happen. It has been found that when Temnothorax albipennis ants are allowed to familiarize themselves with available nest sites before emigrating, this allows for some flexibility in the collective decision to move, by growing an aversion towards sites that are not better than their current nest22. While this aversion may adaptive by allowing them to focus on searching for better alternatives, it can be detrimental when another site is only slightly worse than the current nest, and in a critical time, that aversion slows down emigration. Irrationality in individuals can also arise when cognitive constraints make it difficult to evaluate an option’s actual intrinsic fitness value, and they evaluate an option simply compared to other options. Individual Temnothorax ants, when faced with several complex options where none is clearly superior, make irrational choices, and irrationally change their preference23. The colony, however, seems to have adapted to prevent this irrationality from propagating up to the level of the collective decision, primarily due to its self-organized decentralized decision mechanism described above, minimizing systematic errors arising from individual cognitive constraints. Nevertheless, self-organization is simply not always adaptive. Pushing a parameter value that is important to a system (in this case, decision-making) beyond the range for which the system remains stable can make the selforganized decision inefficient and inaccurate2. Therefore, patterns such as collective decision-making, which are characterized by features such as selforganized decentralized control, are not immune to being maladaptive. They can only be adaptive within a particular range of parameter values, which are not guaranteed to remain within that range. It has been suggested that therefore, individual ant behavior must have been tuned by natural selection to achieve adaptive group behavior2, but recent research suggests otherwise. For example, while it has been claimed that natural selection has pressured collective decisions to be both reactive to change and robust to noise, this has not necessarily been found to be true24. When individuals are selected by the
accuracy of their choices, collective decisions actually don’t turn out optimal! Rather, there evolves a tendency to rely too much on neighbors’ opinions, and the group becomes nearly unresponsive, and when there is a response, it is sudden and rapid. It is, therefore, unclear how adaptive collective decisions really are, and under what conditions this is truly consistent. In order for a selforganized system to be adaptive in an ever-changing environment, it needs a wide range of stable states that is sufficient to react to all perturbations, yet without making its evolution uncontrollably chaotic25. The optimal stable states would theoretically need to be selected by fitness, either directly by a given environment, or by subsystems that have already adapted to that environment, but it is still up to debate whether self-organization is actually adaptive in practice. Adding to this controversy, there is conflicting evidence about the conditions under which self-organization could actually be considered adaptive. Some research suggests that in order for insect foraging to be sufficiently organized to become adaptive, the colony population needs to be above a certain threshold. The Pharaoh’s ant has been shown to phase transition between maladaptive disordered foraging at small populations and adaptive ordered pheromone-based foraging at larger populations26. On the other hand, some evidence suggests that collective decisions work best only in smaller groups. Very recent research has started taking into account the reality of multiple complex correlated spatial and temporal cues, and the infamous “wisdom of the crowd” is not found as often as predicted—meaning, the success of collective decisions does not increase with the size of that collective27. On the contrary, small groups do very well, and it seems to be largely due to noise, which somehow makes it easier to avoid confounding correlated information. Despite all we have learned about collective intelligence, there is much left to re-assess and explore. The landscape of the field is budding with potential for new perspectives and progress. For example, Comparative research into social insect behavior and brain dynamics has led to the idea of “Swarm Cognition”, which is a new approach to studying cognition as a distributed self-organizing pattern28. It has been shown, and used as a tool in much of the research mentioned here, that the algorithmic form of collective decision-making mechanisms can be captured successfully29. Now, there are exciting attempts to harness these algorithms for use in technology—using group food retrieval of ants is used to create multirobot collective transport strategies that share the ants’ decentralized control, and are scalable as well as requiring no previous information about the task at hand, which qualitatively mirrors both individual ant activity and some macroscopic aspects of the group transport30. In the field of medicine, understanding collective decisions may prove useful in studying and treating cancer systems, whose multicellular collective patterns mirror collective decisions in other biological systems31,32. In conclusion, collective intelligence can make group decisions more optimal than individual decisions. A collective decision’s success hinges on numerous conditions, such decentralization, efficient communication, simple
rules and thresholds, amongst others discussed in this paper. There are situations where collective intelligence fails, and the “wisdom of the crowd” is not provably adaptive, and this will require a more directed attention to understand. There is unquestionable potential in the study of this field, and exciting opportunities lay ahead, in fields such as cognition, engineering, and medicine.
References 1.
S. C. Pratt (2010) Collective Intelligence. Encyclopedia of Animal Behavior vol. 1, pp. 303-309. 2. S. Camazine, J. Deneubourg, N. R. Franks, J. Sneyd, G. Theraulaz, E. Bonabeau (2001) Self-Organization in Biological Systems. Princeton University Press. 3. Noam Miller, Simon Garnier, Andrew T. Hartnett, and Iain D. Couzin (2013) Both information and social cohesion determine collective decisions in animal groups. PNAS vol. 110, 13. 5263–5268. 4. Kao AB, Miller N, Torney C, Hartnett A, Couzin ID (2014) Collective Learning and Optimal Consensus Decisions in Social Animal Groups. PLoS Comput Biol 10(8): e1003762. doi:10.1371/journal.pcbi.1003762 5. N. R. Franks, T. O. Richardson, N. Stroeymeyt, R. W. Kirby, W. M. D. Amos, P. M. Hogan, J. A. R. Marshall, T. Schlegel (2013) Speed-cohesion trade-offs in collective decision making in ants and the concept of precision in animal behaviour. Animal Behaviour 85: 1233-1244. 6. Stephen C. Pratt and David J. T. Sumpter (2006) A tunable algorithm for collective decision-making. PNAS.︎ vol. 103, 43. 7. S.C. Pratt (2004) Collective control of the timing and type of comb construction by honey bees (Apis mellifera). Apidologie 35, 193–205. 8. Nigel R. Franks, Stephen C. Pratt, Eamonn B. Mallon, Nicholas F. Britton and David J. T. Sumpter. (2002) Information flow, opinion polling and collective intelligence in house-hunting social insects. Phil. Trans. R. Soc. Lond. DOI 10.1098/rstb.2002.1066 9. S.C. Pratt (2005) Behavioral mechanisms of collective nest-site choice by the ant Temnothorax curvispinosus. Insect. Soc. 52, 383–392. 10. Stephen C. Pratt, Eamonn B. Mallon, David J. T. Sumpter, Nigel R. Franks (2002) Quorum sensing, recruitment, and collective decision-making during colony emigration by the ant Leptothorax albipennis. Behav Ecol Sociobiol 52:117–127. DOI 10.1007/s00265-002-0487-x 11. Zachary Shaffer, Takao Sasaki, Stephen C. Pratt (2013) Linear recruitment leads to allocation and flexibility in collective foraging by ants. Animal Behaviour 86: 967-975. 12. Nils Olav Handegard, Kevin M. Boswell, Christos C. Ioannou, Simon P. Leblanc, Dag B. Tjøstheim, and Iain D. Couzin (2012) The Dynamics of Coordinated Group Hunting and Collective Information Transfer among
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
Schooling Prey. Current Biology 22, 1213–1217. Robinson EJH, Franks NR, Ellis S, Okuda S, Marshall JAR (2011) A Simple Threshold Rule Is Sufficient to Explain Sophisticated Collective Decision- Making. PLoS ONE 6(5): e19981. doi:10.1371/journal.pone.0019981 Sara Brin Rosenthala, Colin R. Twomeyb, Andrew T. Hartnetta, Hai Shan Wub, and Iain D. Couzinb (2015) Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion. PNAS vol. 112 no. 15. Elva J. H. Robinson, Ofer Feinerman, Nigel R. Franks (2012) Experience, corpulence and decision making in ant foraging. Journal of Experimental Biology. 215: 2653-2659; doi: 10.1242/jeb.071076 E. B. Mallon, S. C. Pratt, N. R. Franks (2001) Individual and collective decision-making during nest site selection by the ant Leptothorax albipennis. Behav Ecol Sociobiol, 50:352–359. Naomi E. Leonard, Tian Shen, Benjamin Nabet, Luca Scardovi, Iain D. Couzin, and Simon A. Levin (2012) Decision versus compromise for animal groups in motion. PNAS vol. 109 no. 1, 227–232. Ariana Strandburg-Peshkin, Damien R. Farine, Iain D. Couzin, Margaret C. Crofoot (2015) Shared decision-making drives collective movement in wild baboons. Science Vol. 348 Issue 6241. John R.G Dyer, Anders Johansson, Dirk Helbing, Iain D Couzin and Jens Krause (2009) Leadership, consensus decision making and collective behaviour in humans. Phil. Trans. R. Soc. B, 364, 781-789. doi: 10.1098/rstb.2008.0233 Thomas D. Seeley (2010) Honeybee Democracy. Princeton University Press. (I used "The Five Habits of Highly Effective Honeybees", an extract of chapter 10 of Honeybee Democracy). Couzin, I. D., & Franks, N. R. (2003). Self-organized lane formation and optimized traffic flow in army ants. Proceedings of the Royal Society B: Biological Sciences, 270(1511), 139–146. http://doi.org/10.1098/rspb.2002.2210 Nathalie Stroeymeyt, Elva J.H. Robinson, Patrick M. Hogan, James A.R. Marshall, Martin Giurfa,and Nigel R. Franks (2011) Experience-dependent flexibility in collective decision making by house-hunting ants. Behav Ecol 22:535–542. Susan C. Edwards, and Stephen C. Pratt (2009) Rationality in collective decision-making by ant colonies. Proc. R. Soc. B 276, 3655–3661. doi:10.1098/rspb.2009.0981 Torney CJ, Lorenzi T, Couzin ID, Levin SA. (2015) Social information use and the evolution of unresponsiveness in collective systems. J. R. Soc. Interface 12: 20140893. http://dx.doi.org/10.1098/rsif.2014.0893 Francis Heylighen (2014) The Science Of Self- Organization And Adaptivity. Knowledge Management, Organizational Intelligence And
26.
27.
28.
29.
30.
31. 32.
Learning, And Complexity, Vol. I. Madeleine Beekman, David J. T. Sumpter, and Francis L. W. Ratnieks (2001) Phase transition between disordered and ordered foraging in Pharaoh's ants., vol. 98 no. 17 , 9703–9706, doi: 10.1073/pnas.161285298 Kao AB, Couzin ID. 2014 Decision accuracy in complex environments is often maximized by small group sizes. Proc. R. Soc. B 281: 20133305. http://dx.doi.org/10.1098/rspb.2013.3305 Vito Trianni, Elio Tuci, Kevin M. Passino, James A.R. Marshall (2011) Swarm Cognition: an interdisciplinary approach to the study of selforganising biological collectives. Swarm Intell 5: 3–18. DOI 10.1007/s11721-010-0050-8 Stephen C. Pratt, David J. T. Sumpter, Eamonn B. Mallon & Nigel R. Franks (2005) An agent-based model of collective nest choice by the ant Temnothorax albipennis. Animal Behaviour, 70, 1023–1036. Spring Berman, Quentin Lindsey, Mahmut Selman Sakar, Vijay Kumar, and Stephen C. Pratt (2011) Experimental Study and Modeling of Group Retrieval in Ants as an Approach to Collective Transport in Swarm Robotic Systems. Proceedings of the IEEE, Vol. 99, No. 9. Thomas S. Deisboeck and Iain D. Couzin (2009) Collective behavior in cancer cell populations. BioEssays 31:190–197. Woods ML, Carmona-Fontaine C, Barnes CP, Couzin ID, Mayor R, et al. (2014) Directional Collective Cell Migration Emerges as a Property of Cell Interactions. PLoS ONE 9(9): e104969. doi:10.1371/journal.pone.0104969
