Nature of collective decision-making by simple yes/no decision units

Abstract

The study of collective decision-making spans various fields such as brain and behavioural sciences, economics, management sciences, and artificial intelligence. Despite these interdisciplinary applications, little is known regarding how a group of simple ‘yes/no’ units, such as neurons in the brain, can select the best option among multiple options. One prerequisite for achieving such correct choices by the brain is correct evaluation of relative option quality, which enables a collective decision maker to efficiently choose the best option. Here, we applied a sensory discrimination mechanism using yes/no units with differential thresholds to a model for making a collective choice among multiple options. The performance corresponding to the correct choice was shown to be affected by various parameters. High performance can be achieved by tuning the threshold distribution with the options’ quality distribution. The number of yes/no units allocated to each option and its variability profoundly affects performance. When this variability is large, a quorum decision becomes superior to a majority decision under some conditions. The general features of this collective decision-making by a group of simple yes/no units revealed in this study suggest that this mechanism may be useful in applications across various fields.

Adaptive phenotypic variation among clonal ant workers

“…Another possibility is that the optimal threshold distribution is different between the populations. Recently, it has been shown that the performance of collective decision by using only yes/no decision units is maximized when the threshold distribution of a collective decision-maker is matched with the distribution of the options' qualities [ 22 ]. This issue is worth examining in future studies.…”

What Could Models of Superorganismal Cognition Offer to Embodied AI?

“…Third, biological research points toward the value of heterogeneity in behavioural thresholds for more accurate and resilient collective decisions [21,22]. For instance, a small proportion (3-12%) of Temnothorax workers are persistent in exploring during a nest-site decision [23], which may correspond to high acceptance threshold 'picky' ants who keep looking for the very highest quality nest sites.…”

Response Threshold Distributions to Improve Best-of-N Decisions in Minimalistic Robot Swarms

“…which response threshold distribution D h ) maximise the probability of success given a known number of options n and distribution of options' quality D q . Past experiments only considered Gaussian distributions of the options' qualities [9,35,24], however, environments with other distributions may exist. In our analysis, we consider three types of quality distributions D q ∈ {U, N , K}: the uniform distribution U(µ q , σ q ), the Gaussian distribution N (µ q , σ q ), and the bimodal distribution K(µ ′ q , µ ′′ q , σ ′ q , σ ′′ q ).…”

“…Behaviours based on response thresholds have a reactive binary (yes/no) response determined by the stimulus intensity (or option quality) being above or below a threshold, and can be observed in several eusocial insect species [8,24,18,25,34,27]. Despite individuals' simplicity, systems composed of response threshold units can display accurate and rational collective behaviour [24,35,9,17], which is enabled by a crucial element of such systems, the heterogeneity of their individuals, each having a different threshold [18,8].…”

“…The division of labour in ants, regulated by heterogeneous response thresholds, has inspired the design of several multirobot systems to tackle task allocation problems [13,1,15,16,6,26,12,4]. However, despite its potential, only limited attention has been devoted to apply response thresholds to the design of consensus decision making systems [9,23]. While the long term objective is the deployment of minimalistic binary robots for best-of-n decisions, this paper only focuses on the relationship between threshold distribution and environmental stimuli in order to improve the collective ability to distinguish between options and select the best.…”