Abstract
We develop a model of information exchange through communication and investigate its implications for information aggregation in large societies. An underlying state determines payoffs from different actions. Agents decide which others to form a costly communication link with, incurring the associated cost. After receiving a private signal correlated with the underlying state, they exchange information over the induced communication network until taking an irreversible action. We define asymptotic learning as the fraction of agents taking the correct action converging to one in probability as a society grows large. Under truthful communication, we show that asymptotic learning occurs if most agents are a short distance away from "information hubs," which receive and distribute a large amount of information. Asymptotic learning therefore requires information to be aggregated in the hands of a few agents. We also show that while truthful communication may not always be a best response, it is an equilibrium when the communication network induces asymptotic learning. Moreover, we contrast equilibrium behavior with a socially optimal strategy profile. We show that when the network induces asymptotic learning, equilibrium behavior leads to maximum aggregate welfare, but this may not be the case when asymptotic learning does not occur. Finally, we demonstrate that societies with too many and sufficiently large social cliques do not induce asymptotic learning, as self-sufficiency within cliques discourages broader communication.
Introduction
In an era characterized by fragmented media landscapes and the rapid spread of information, understanding how societies reach a consensus on the truth is a critical challenge. Traditional models of social learning often assume that communication networks are fixed or that the exchange of information is costless. However, in reality, individuals choose whom to listen to, and acquiring information requires time and resources.
This paper investigates the dynamics of information exchange within endogenous social networks—networks where the links are formed based on the strategic decisions of agents. The central question is whether a large society can successfully aggregate noisy, dispersed signals into a perfect collective understanding, a phenomenon termed "asymptotic learning."
The Model of Endogenous Exchange
The authors propose a game-theoretic framework consisting of two distinct stages:
- Network Formation: Agents decide which other individuals to form directed communication links with. Each link carries a specific cost. This stage recognizes that attention and connection are finite resources.
- Information Exchange: Each agent receives a private, noisy signal regarding the state of the world. Information is then exchanged over the network formed in Stage 1. This process is stochastic, modeled via a Poisson process, and involves a "cost of time" (discounting), meaning agents must balance the desire for more accurate information with the need to act quickly.
Using Bayesian updating with Normal distributions, the model tracks how beliefs evolve as information travels through the network. The analysis identifies two pivotal roles within these networks: Information Mavens (agents with high in-degree who aggregate data) and Social Connectors (agents with high out-degree who disseminate data).
The Necessity of Information Hubs
A primary finding of the research is that asymptotic learning is inextricably linked to the emergence of information hubs. Contrary to some theories suggesting that influential agents create "herds" that stifle independent thought, this model demonstrates that hubs are essential for aggregation. For a society to learn the truth as it grows larger, information must be concentrated in the hands of a few (mavens) and then redistributed efficiently (connectors).
Specifically, asymptotic learning occurs if and only if the vast majority of agents are within a short distance of these hubs. Without this concentration and subsequent redistribution, information remains too dispersed to overcome the noise of individual signals.
The "Social Clique" Trap and Informational Braess' Paradox
One of the most counterintuitive insights of the study is the role of "social cliques"—groups of individuals who can communicate at zero or very low cost (e.g., close friendship circles). While local connectivity seems beneficial, the authors show that if these cliques are too large or too numerous, they can actually prevent the society from reaching a global consensus.
When a clique is large enough to provide "good enough" information for free, its members lose the incentive to pay for links to agents outside their circle. This leads to informational segregation. This phenomenon is a variation of Braess' Paradox: adding information to a local group can decrease total societal welfare. By making a local group more self-sufficient, you destroy the communication paths that other, less-informed agents relied on to access broader information.
Equilibrium, Truthfulness, and Efficiency
The model also addresses the strategic nature of communication. While agents might theoretically have incentives to misreport information to manipulate the timing of their neighbors' actions, the authors prove that when a network structure is conducive to asymptotic learning, truthful communication is a Nash equilibrium.
Furthermore, the paper explores the gap between decentralized equilibrium and social optimality. In scenarios where asymptotic learning occurs, the decentralized choices of individual agents align with what a social planner would choose to maximize total welfare. However, when learning fails—often due to the aforementioned clique trap—the equilibrium behavior is no longer socially efficient.
Conclusion and Future Directions
The research highlights that the structure of our social networks is a double-edged sword. While hierarchical structures like pyramids are highly effective at inducing learning through the creation of mavens and connectors, the presence of large, self-sufficient cliques can lead to fragmentation and the failure of information aggregation.
Future research may expand on these findings by relaxing the assumption of homogeneous preferences. In a polarized environment where agents have conflicting interests, the incentives for strategic lying may become more pronounced. Additionally, moving from a static network formation to a dynamic one—where links can be adjusted as information arrives—would provide further insight into the resilience of social learning in a changing world.
Summary updated