SOCIAL RECOMMENDATIONS RATHER THAN SOCIAL VALUES

摘要

Two different modes of social influence on human or artificial decision makers operating in complex environments are studied. One is through emulation of social values, and the other is through social recommendations for specific actions. Two armed bandit problems are examined where agents are influenced by their peer group's attitude. It is observed that emulation of social values may speed up learning, but may also coordinate agents on a sub-optimal mode of behavior. Social recommendations on the other hand, do help in learning equilibrium play in the same context. The results of the paper shed some light on how to incorporate social influence in stochastic decision environments.