| Description |
Abstract:
Collective decision-making can be more accurate than individual decision-making, due to the ability of groups to aggregate information from many individuals. However, information aggregation is vulnerable to a variety of social and cognitive biases which can undermine the "wisdom of the crowds" and result in bad decisions. Here, we study the effect of two very hard-to-avoid kinds of biases: the tendency of similar individuals to be connected, and the tendency to share some kinds of information more than others. We build mathematical models of a society composed of rational agents that use Bayesian inference to learn about the world, combining their direct observations with information they obtain from other agents. We include two forms of transmission bias: (1) assortment in communication networks, where individuals preferentially connect with similar others, and (2) selective sharing of signals. We identify conditions under which these effects are sufficient to generate stable divergence of beliefs across the population--either in the form of polarization or the entire population being misled. We show that when the population is balanced between types of agents, the collective is most sensitive to the effects of assortment. This is also the condition when the collective is most responsive to the environment. Therefore, belief polarization is most likely under conditions that support the greatest collective ability to sense the truth. Our results thus identify tradeoffs between accuracy and consensus in collective decision-making and highlight structural limits to collective intelligence, even in an idealized setting. |
QLS-TLQS Seminar: Biased Transmission Drives an Accuracy-Consensus Tradeoff in Collective Decision-making
Go to day