Since the introduction of economic theory to political science, theorists have argued that discussion could serve as an effective information shortcut if individuals communicate with experts who have similar preferences. Previous experimental and survey studies have found mixed results for the efficacy of social communication, but they have not observed the process of discussion partner selection which is so central to the previous models. This paper presents the results of a group-based experiment that allows for discussion partner selection. We fail to find aggregate enlightenment through social communication: lesser informed subjects are helped by social communication, but better informed subjects are harmed. This result is caused in part because subjects are too willing to seek out more expert discussion partners who have different ex ante preferences.