The Emergence of Cultural Attractors: How Dynamic Populations of Learners Achieve Collective Cognitive Alignment

Abstract

Cultural attractor landscapes describe the time-evolution of cultural variants (i.e. behaviors, artifacts) over successive transmission events. When cultural variants sit at a local minimum of a stable attractor landscape, there will be no cumulative error over transmissions, allowing Darwinian selectionist dynamics to drive cumulative cultural evolution. But because cultural attractors are the emergent products of shifting populations (individuals can leave or die, and new individuals can be born or enter the population) of individuals with potentially-idiosyncratic and dynamic cognitive landscapes, which are in turn the emergent products of individual experience within a culture, stable cultural attractor landscapes cannot be taken for granted. Little is known about how cultural attractors form or stabilize in light of this culture-cognition feedback loop. We present a model of cultural attractor dynamics, which adapts a cognitive model of unsupervised learning of phoneme categories in individual learners to a multi-agent setting wherein individual learners are tasked with categorizing and generating signals for one another. Beginning from a state in which all agents possess a set of randomly distributed categories of uniform probability, under some conditions populations self-organize into signal clusters, which constitute an identifiable cultural attractor landscape. We explore the role of various innate cognitive biases, levels of transmission error, learning periods, lifespans, population sizes, and network structures to understand when population-level structure may emerge, what properties it is likely to have, and how stable it is. Our analyses can provide insights into the conditions that may be favorable or unfavorable for cumulative culture to emerge.