Manipulating Opinions in Social Networks

We consider a general model of boundedly rational opinion formation in social networks. We show that long run opinions are extremely vulnerable to unilateral subtle manipulation. For a given updating system, any agent can drive the long run opinions of all agents to an arbitrary desired opinion, by infinitesimally perturbing his updating behavior. However, when opinion formation is monitored, then as the perturbation goes to zero so does the extend to which long run opinions can be manipulated without detection. Finally, we show that asymptotic consensus is a robust outcome of boundedly rational opinion formation while naive learning and social influence are not.