Messengers: Breaking Echo Chambers in Collective Opinion Dynamics with Homophily

Abstract: Collective estimation is a variant of collective decision-making, where agents need to achieve consensus on a continuous quantity in a self-organized fashion via social interactions. A particularly challenging scenario is a fully distributed collective estimation with strongly constrained, dynamical interaction networks, for example, encountered in real physical space. Collectives face several challenges in achieving precise estimation consensus, particularly due to complex behaviors emerging from the simultaneous evolution of the agents' opinions and the interaction network.While homophilic networks may facilitate collective estimation in well-connected networks, we show that disproportionate interactions with like-minded neighbors lead to the emergence of echo chambers, preventing collective consensus. Our agent-based simulation results confirm that, besides a lack of exposure to attitude-challenging opinions, seeking reaffirming information entraps agents in echo chambers. In a potential solution, agents can break free from the pull of echo chambers. We suggest an additional state where stubborn mobile agents (called Messengers) carry data and connect the disconnected clusters by physically transporting their opinions to other clusters to inform and direct the other agents. However, an agent requires a switching mechanism to determine which state to adopt. We propose a generic, novel approach based on a Dichotomous Markov Process (DMP). We show that a wide range of collective behaviors arise from the DMP. We study a continuum between task specialization with no switching (full-time Messengers), generalization with slow switching (part-time Messengers), and rapid task switching (short-time Messengers). Our results show that stubborn agents can, in various ways, help the collective escape local minima, break the echo chambers, and promote consensus in collective opinion dynamics.