Exploring the effect of network structure on individual learning, a longitudinal study of an online Go game community

Culture is a population phenomenon that emerges as a consequence of the exchange of information between individuals. For this reason it is assumed that the properties of these networks have an effect on the processes of cultural evolution and accumulation. Some general hypotheses have been evaluated, but testing more specific ones has been hampered by the need for detail data on the behavior of all individuals in a community over time. To explore how the structure of a information exchange network affects human learning, we studied over 8 years the evolution of the Go player network of a major online service platform. Because drawing causal conclusions from observational data requires specifying a causal model, based on the main findings of the literature on cumulative cultural evolution, we state that human learning depends on at least four factors: environment, individual experience, cultural information and initial ability. Knowing the initial skills is fundamental because, in all structural causal models, the learning rate depends on the stage in which the individuals are before starting the process. Thanks to our implementation of the state-of-the-art skill model, we obtained good estimates throughout the entire time series, even at the beginning, allowing us to compare novice, intermediate and expert individuals separately. In addition, the causal model assumes that access to cultural information depends on the information exchange network. To explore the causal effect that the network has on individual learning, we use classical dynamic graph analysis techniques to determine the topological position that individuals have in the network over time. We find that moderate values of centrality have a positive second-order effect only on players who are in the middle of the learning process. Novice players learn mainly by individual experience, and expert players have nothing more to learn. To the best of our knowledge, this is the first work that explores the effect that position in a dynamic network has on individual learning.