Learning from neighbours

Abstract

Agents navigating environments with unknown payoffs utilize both personal experience and the observed outcomes of neighbors to update their beliefs and maximize utility. Within any connected society, this localized learning process ensures that all agents eventually achieve identical expected payoffs, leading to social conformism when actions have distinct rewards. The likelihood that a society adopts the optimal action depends heavily on the specific architecture of its neighborhood networks. High connectivity does not always facilitate efficiency; centralized information sources, such as a universally observed “royal family,” can propagate negative signals that prematurely terminate experimentation with superior technologies. In contrast, the presence of locally independent agents—those possessing non-overlapping neighborhoods—promotes the generation of diverse information and increases the probability of global adoption of the optimal choice. Mathematical proofs and simulations demonstrate that these dynamics produce adoption rates following a logistic function and spatial diffusion patterns consistent with empirical findings in technology and agricultural innovation. – AI-generated abstract.