Can human moderators ever really rein in harmful online content? New research says yes