This project examines how animals in groups collectively learn about their environment and how this learning process may result in "collective intelligence", whereby the group can make more accurate decisions than any single individual can. Most social animals, such as fish, have very limited means of communication and coordination, and it is currently unknown whether these animals can learn to exploit the potential collective intelligence under these constraints. This project will develop a theoretical model of collective learning, and preliminary results suggest that the same learning rules known to be ubiquitous in solitary animals can also be highly efficient in a group context and can lead to near-optimal levels of collective intelligence, even with limited coordination between individuals. The specific predictions of the collective learning model will then be experimentally tested using two social animal species: fish and humans. By using two very different species, the generality of the model can be tested, and general mechanisms of collective learning will be uncovered. These experiments will provide important insights into the learning process in animal groups, which has previously been unexplored. They may also give specific insight into how humans learn in a group context and may suggest techniques to increase the efficiency of collective learning in human groups.
In this project, we studied how animals learn within a social context, and the impact that collective learning has on how well animal groups make decisions together, a phenomenon known as 'the wisdom of crowds.' Because many animal groups are highly cohesive and make consensus decisions, the preferences of some group members may greatly impact what an individual experiences in the environment and therefore learns about. Through a series of computational models and experiments, we probed the assumptions of our model, as well as extensions of the model. We discovered that using minimal assumptions, individuals are capable of learning behavior that maximizes collective intelligence, due to the feedback between consensus decisions and individual learning. Therefore, even with limited knowledge of their environment and the preferences of group members, individuals can readily and robustly learn behavior that maximizes performance in a collective context. We also found that in cases in which learning is not possible (which might be the case where environments change too rapidly, or organisms that have relatively fixed behavior, for example), small group sizes, rather than the large groups we tend to assume, make the most accurate decisions. Related to this, internal structure in group, such as hierarchical structures, tend to decrease the 'effective group size,' allowing large groups to behave like smaller groups and therefore improve their decision accuracy. This is due to a tension between the detrimental effects of correlated information, which larger groups are more susceptible to, and the positive effects of pooling independent estimates. We find that small to medium size groups strike a balance between these two effects. These findings modify substantially our predictions for animal group sizes in nature. In addition to these computational models, we performed experiments on three species: zebrafish, humans, and the slime mold Physarum polycephalum, in order to investigate the generality of the results of the computational models in a wide variety of biological organisms. We have gained an understanding of when zebrafish do, or do not, reach consensus during decision bouts, how humans' opinions are affected by social information, and how slime molds integrate information about their local environment across the entire network. By finding commonalities as well as difference between these model organisms, we can begin to build a more comprehensive understanding of collective decision-making across biological taxa and scales. Together, these findings substantially improve our understanding of how distributed organisms or groups can make collectively intelligent decisions. We have disseminated our findings through peer-reviewed scientific publications, conference talks, through directly interacting with the public during our human experiment, and through fostering collaborations between researchers and undergraduates from very different scientific backgrounds. Using these multiple channels, we hope to improve progress in our interdisciplinary research topic, as well as improve the public's understanding of research on the wisdom of crowds in particular, and science in general.
