(A) simulation study on the efficiency-increasing effect of non-selective imitation in a multi-agent system

Abstract

Imitation, the act of copying the behavior of others, is widely observed in social animals. It has been suggested that imitation can induce cultural evolution but only when imitation targets are selectively chosen (i.e., biased imitation). On the other hand, it has also been proposed that random copying (i.e., unbiased imitation) is observed in human subject experiments and can describe certain cultural phenomena. Nonetheless, a detailed mechanism of how random copying can induce population-level learning remains elusive. Here, using a multi-agent survival model, we show that random copying can function as a key mechanism of population evolution, in particular, increasing the survival time of agents. When agents were only under the effect of width mutations, the average lifespan did not change over time. However, surprisingly, the average lifespan increased over time when the agents were allowed to randomly imitate each other. Moreover, their trait converged to the optimal trait in the given environment, maximizing their lifespan. Next, we found that random copying groups reache comparable or even higher average lifespans than selective copying groups. This implies that random copying can potentially serve as a better learning strategy than selective copying. Lastly, we examined the effect of random copying by varying the duration and timing of the random copying. Interestingly, the increase in the average lifespan was maximized when each agent performed imitations only at the early stage of their life. In summary, our results suggest that random copying can function as an effective learning method and have an important role in the evolution in a multi-agent system.