Conformist social learning leads to self-organised prevention against adverse bias in risky decision making

Abstract

Given the ubiquity of potentially suboptimal biases widely observed in animal behaviour, it seems paradoxical that improvements of decision accuracy through the conformist-based social learning is still widely observed. Here we show, through model analyses and online experiments with 467 adult human subjects, that the frequency-based copying can promote adaptive risk-taking in repeated decision-making, even though many individuals are potentially biased toward suboptimal risk-aversion due to myopia of reinforcement learning. Although a positive feedback imposed by conformity could result in maladaptive informational cascades, our behavioural dynamical model identified a key role of negative feedback that arises when the inherent behavioural bias is relaxed by a weak minority influence. This 'collective behavioural rescue', emerging through coordination of positive and negative feedback, highlighted a benefit of social learning in wider environmental conditions than previously assumed, and resolved the ostensible paradox of flexible conformity under risk and uncertainty.