Individual Learning and Collective Intelligence∗

In this paper, we explore how incentives and information structures influence emergent collective accuracy. In our model, individuals of bounded ability predict outcomes that depend on the values of a set of attributes. Each individual constructs a predictive model that considers only a subset of those attributes. We compare environments in which individuals learn in isolation to environments where they can learn from and copy more successful predictors. We also compare incentives for accurate individual performance to market incentives, in which agents are paid an amount proportional to the inverse of the percentage of agents who predict correctly. We find that market incentives and isolated learning environments produce less accurate but more diverse predictors and that the benefits of diversity outweigh the loss in accuracy. Paradoxically, the incentives and information structures that produce less accurate individuals produce more accurate collectives.