The Repeatability of Human Swarms

Abstract

Swarm Intelligence (SI) is a natural phenomenon in which social organisms amplify their decision-making abilities by forming real-time systems that converge on optimized solutions. It has been studied extensively in schools of fish, flocks of birds, and swarms of bees. In recent years, a new technology called Artificial Swarm Intelligence (ASI) has enabled human groups to form similar systems over computer networks. While “human swarms” have been shown to be more accurate than traditional methods for tapping the intelligence of human groups, the present study tests the repeatability of the answers that human swarms generate. Ten groups of 20 to 25 participants were asked to give subjective ratings on a set of 25 opinion-based questions. The groups answered by working together in real-time, connected by swarming algorithms. The results show that groups answering as swarms produce repeatable results, reaching the same answer as other groups 67% of the time. Additional analysis found that the repeatability of each swarm was significantly correlated with a Conviction Index (CI) metric computed from the real-time swarming data (r2 = 0.33, p < 0.01). For swarms that converged upon a solution with a Conviction Index (CI) > 85%, the repeatability was found to be greater than 90% and the likelihood that another swarm randomly sampled from a similar population would generate the same response was greater than 95% (p < 0.05). This provides powerful guidelines for groups using ASI technology to generate optimized forecasts, insights, and decisions from human swarms sampled from general populations.