Who should fight the spread of fake news?

TL;DR

It is found that debunking in certain circumstances can be a counterproductive strategy, whereas some targeted strategies (akin to"deplatforming") and/or preemptive campaigns turn out to be quite effective.

Abstract

This study investigates who should bear the responsibility of combating the spread of misinformation in social networks. Should that be the online platforms or their users? Should that be done by debunking the "fake news" already in circulation or by investing in preemptive efforts to prevent their diffusion altogether? We seek to answer such questions in a stylized opinion dynamics framework, where agents in a network aggregate the information they receive from peers and/or from influential external sources, with the aim of learning a ground truth among a set of competing hypotheses. In most cases, we find centralized sources to be more effective at combating misinformation than distributed ones, suggesting that online platforms should play an active role in the fight against fake news. In line with literature on the "backfire effect", we find that debunking in certain circumstances can be a counterproductive strategy, whereas some targeted strategies (akin to "deplatforming") and/or preemptive campaigns turn out to be quite effective. Despite its simplicity, our model provides useful guidelines that could inform the ongoing debate on online disinformation and the best ways to limit its damaging effects.

Figures (12)

• Figure 1: Temporal evolution (left) of Truthfulness, (right) of CD, under 'mega-nodes.' Results are obtained by averaging over $100$ independent simulations run on sparse ER networks with $N=100$ and $M=4$. Legend: $mg_a$ and $mg_b$ refers to conspiring and debunking mega-node respectively where $0$ denotes absence and $1$ presence
• Figure 2: Left: Temporal evolution of truthfulness when disinformation originates from a centralized source (mega-node) or a sub-population of distributed conspirator agents amounting to a fraction $\beta_c$ of the whole population. Right: Temporal evolution of truthfulness in the presence of debunking from both centralized (one mega-node, red line) and decentralized (a fraction $\beta_d$ of the population) sources against disinformation originated from a decentralized source (a fraction $\beta_c$ of the population). Results are obtained by averaging $100$ independent simulations run on sparse ER networks with $N=100$ and $M=4$.
• Figure 3: Left: Temporal evolution of CD when disinformation originates from a centralized source (mega-node) or a sub-population of distributed conspirator agents amounting to a fraction $\beta_c$ of the whole population. Right: Temporal evolution of CD in the presence of debunking from both centralized (one mega-node, red line) and decentralized (a fraction $\beta_d$ of the population) sources against disinformation originated from a decentralized source (a fraction $\beta_c$ of the population). Results are obtained by averaging $100$ independent simulations run on sparse ER networks with $N=100$ and $M=4$.
• Figure 4: Left: Temporal evolution of truthfulness in scenarios with decentralized debunking and prebunking, carried out by a fraction $\beta_d$ and $\beta_p$ of the population, respectively. Right: steady-state truthfulness obtained in simulations with different concentrations $\beta_p$ of decentralized prebunker agents as a function of the concentration $\beta_d$ of debunker agents where errorbars capture the differences across simulations. Results are obtained by averaging $100$ independent simulations run on sparse ER networks with $N=100$ and $M=4$.
• Figure 5: Temporal evolution of truthfulness in scenarios with/without centralized debunking and/or prebunking. Results are obtained by averaging $100$ independent simulations run on sparse ER networks with $N=100$ and $M=4$.