Several forms of social learning rely on the direct or indirect evaluation of the fitness of cultural traits. Here we argue, via a simple agent-based model, that payoff uncertainty, that is, the correlation between a trait and the signal used to evaluate its fitness, plays a pivotal role in the spread of beneficial innovation. More specifically, we examine how this correlation affects the evolutionary dynamics of different forms of social learning and how each form can generate divergent historical trajectories depending on the size of the sample pool. In particular, we demonstrate that social learning by copying the best model is particularly susceptible to a sampling effect caused by the interaction of payoff uncertainty, the number of models sampled (the sample pool), and the frequency with which a trait is present in the population. As a result, we identify circumstances in which smaller sample pools can act as "cultural incubators" that promote the spread of innovations, while more widespread sampling of the population actually retards the rate of cultural evolution.
Keywords: agentbased model; cultural transmission; payoff uncertainty; payoff-biased transmission; spread of innovation.