Evaluating Social Learning Strategies:
The Impact of Skill Dependencies and Meta-Learning in an Agent-Based Model

Research Master Psychology Internship at the University of Amsterdam

Abstract: Learning skills is tremendously important for humans. Social learning enables humans to transmit skills between individuals and across generations. Previous studies investigated different learning strategies but were limited in two ways: agents mostly applied one learning strategy and skills were perceived as independent from each other. We addressed these gaps with an agent-based model that investigates the efficacy of four social learning strategies (success-, similarity-, age-, conformity-based) and three types of meta-learning (fixed, flexible, integrative) in two skill trees (independent skills and linear dependence of skills). In a total of 16 simulations, we found that the level of dependency between skills influenced the performance of learning strategies and meta-learning types. While we found no differences between learning strategies and meta-learning types when skills were independent from each other, the similarity- and age-based strategy outperformed the other strategies when skills were linearly dependent. These findings propose a starting point for future research to further evaluate how constraints and dependencies between skills can influence the performance of learning strategies. Further, our exploratory research into meta-learning revealed promising theories about what learning of social learning strategies could look like. Keywords: Social Learning, Meta-Learning, Skill Learning