Hyperswarms: A New Architecture for Amplifying Collective Intelligence

Artificial Swarm Intelligence (ASI) is a recently developed method that enables networked human groups to converge on more accurate group forecasts, estimations, and decisions. While ASI can significantly amplify collective intelligence, the process struggles when too large of a majority supports an inaccurate view, even if their average confidence is low. Thus, a major goal of ASI research is to increase resilience to low-confidence majorities. This paper introduces a new ASI structure called a Hyperswarm that enables a confident minority to more readily sway an unsure majority. The approach involves dividing a population P into a set of overlapping sub-groups (H
1
, H
2
… H
P
) such that each member only interacts with members of their subgroup. And because each subgroup overlaps multiple other subgroups, the local interactions quickly propagate throughout the full population. In this paper we simulate hyperswarms, showing that a confident minority can intelligently overcome a less confident majority, even when 70% of participants initially harbor the majority view. In addition, we explore a variety of hyperswarm design parameters and derive guidelines for future development.