Collective Decision Making in Distributed Systems

We propose cognitive design patterns (CDPs) as a general methodology to design distributed artificial systems capable of cognitive processing. CDPs are reusable solutions that leverage the current understanding of cognitive processing in natural distributed systems, and that put this knowledge in use for the design of artificial ones [2]. In this paper, we derive a CDP for collective decision making inspired by the nest-site selection behaviour of honeybee swarms [1]. During this process, honeybees leave the swarm in search of a new nest site. A bee that discovers a potential nest gets committed to it and returns to the swarm to recruit other uncommitted scouts through the waggle dance. Bees committed to different alternatives deliver to each-other stop signals, which make them return to a uncommitted state. Additionally, scouts have a certain probability of spontaneously abandoning commitment. This collective process is based on peer-to-peer interactions among bees, and leads to a consensus decision for the best available alternative. Additionally, thanks to the cross-inhibition mechanism implemented through stop signals, the colony is able to break decision deadlocks in case of same-quality alternatives, and to randomly selects any of the two. An analytical model of the nest-site selection process in a binary-choice scenario has been developed and confronted with empirical results [1]. The model is a ODE system that describes the dynamics of the fraction Ψi = Ni/N of individuals belonging to population i ∈ {U,A,B}, where U is the population of uncommitted scouts, andA andB are the populations of bees committed to option A or B. The model assumes that agents switch between populations through four types of transitions with constant rate: discovery (γi), abandonment (αi), recruitment (ρi) and cross-inhibition (σi):