Quantifying effects of tasks on group performance in social learning

Social learning is a learning process in which new behaviors can be acquired by observing and imitating others. It is the key to cultural evolution because individuals can exchange profitable information culturally within the group. Recent studies have over-focused on social learning strategies but paid rare attention to the learning tasks. In particular, in these studies, individuals rely on perfect imitation, directly copying the solutions of others, to improve their performance. However, imperfect imitation, a prevalent form of social learning in cultural evolution, has received little discussion. In this paper, the effects of three task features (task types, task complexity, and task granularity) on group performance are simulated with an agent-based model and quantified with decision trees. In the proposed model, individuals in a network learn from others via imperfect imitation, which means individuals make a trade-off between their solutions and socially acquired solutions. Here, status quo bias is introduced to represent the degree to which individuals adhere to their solutions. Results show that the performance of a group is not affected by task complexity in hard-to-easy tasks but declines with the task complexity rising in easy-to-hard tasks. Besides, groups usually perform better in fine-grained tasks than in coarse-grained ones. The main reason is that in coarse-grained tasks, conservative individuals encounter learning bottlenecks that prevent them from exploring superior solutions further. Interestingly, increasing task granularity can mitigate this disadvantage for conservative individuals. Most strikingly, the importance scores given by decision trees suggest that tasks play a decisive role in social learning. These findings provide new insights into social learning and have broad implications for cultural evolution.