How fake news can turn against its spreader

1. Problem Statement

The paper addresses the growing impact of disinformation ("fake news") on democratic processes and public perception. While existing literature often focuses on detecting or debunking fake news, this study investigates a counter-intuitive phenomenon: under specific conditions, the dissemination of disinformation can backfire, producing an effect opposite to the spreader's intent.

The authors aim to model how rational receivers process multiple information sources when one or more contain hidden biases (disinformation). They seek to determine if and when the combination of genuine and fake information leads to a "reversal effect," where the receiver's updated belief moves away from the disinformation spreader's desired outcome.

2. Methodology

The authors employ a framework rooted in Communication Theory and Bayesian Inference, specifically utilizing Information-Based Asset Pricing models adapted for social dynamics.

• Signal-plus-Noise Model: Information is modeled as a random variable $X$ (the true state, e.g., a candidate's position) corrupted by noise $\epsilon$. The received message is $\xi = \sigma X + \epsilon$, where $\sigma$ is the signal-to-noise ratio (SNR).
• Disinformation Modeling: Disinformation is modeled as an additive bias $f$ in the noise term, such that the message becomes $\xi = \sigma X + (\epsilon + f)$. The receiver is assumed to be unaware of $f$, treating the biased message as if it were pure noise.
• Bayesian Updating: The receiver updates their prior probabilities $\{p_k\}$ regarding the state $X$ using Bayes' formula upon receiving $\xi$.
• Information Fusion: The core methodology involves combining two distinct information sources ($\xi_1$ and $\xi_2$) into a single effective information source. The authors derive the mathematical conditions under which the noise terms of these sources are correlated ($\rho$).
• Continuous Time Extension: The model is generalized to a time-series setting using Brownian motion to simulate the continuous arrival of information over an election cycle.
• Overshoot Analysis: The paper analyzes the non-linear relationship between the magnitude of disinformation and the resulting posterior probability, specifically looking at intermediate candidates on a political spectrum.

3. Key Contributions

The paper makes three primary theoretical contributions:

1. The Reversal Effect (Backfiring): The authors prove that when two information sources are combined, if the noise terms are positively correlated ($\rho > 0$) and the signal-to-noise ratio of the genuine source ($\sigma_1$) is sufficiently large relative to the disinformation source ($\sigma_2$), the disinformation can generate an opposite effect.
   • Condition: If $\sigma_2 < \rho \sigma_1$, a positive bias $f$ intended to shift belief toward higher values of $X$ will instead shift belief toward lower values.
2. The Overshoot Effect: The paper identifies a scenario where excessive disinformation pushes the receiver's belief past the intended target. If a candidate attempts to shift voters toward an extreme position using disinformation, an excessive amount of such noise can cause the receiver to skip the intended intermediate candidate and jump to the most extreme candidate on the spectrum.
3. Strategic Implication for Countering Disinformation: The study suggests that the most effective counter-strategy is not necessarily fact-checking (which is retrospective) but rather increasing the signal strength of genuine information. By raising the SNR of truthful sources, the impact of correlated noise (disinformation) can be mathematically neutralized or reversed.

4. Key Results

A. The Reversal Mechanism

When a receiver consumes two messages:
$$\xi_1 = \sigma_1 X + \epsilon_1$$
$$\xi_2 = \sigma_2 X + \epsilon_2 + f_2$$
(where $\epsilon_1, \epsilon_2$ are correlated with correlation $\rho$), the combined effective message results in a shift of the signal $X$.

The authors derive that the net shift in the receiver's perception is proportional to:
$$\Delta X \propto \frac{\sigma_1 - \rho \sigma_2}{\sigma^2(1-\rho^2)} f_1 + \frac{\sigma_2 - \rho \sigma_1}{\sigma^2(1-\rho^2)} f_2$$

• The Critical Finding: If $\rho > 0$ and $\sigma_2 < \rho \sigma_1$, the coefficient for $f_2$ becomes negative.
• Interpretation: If a right-leaning party (intending to push voters to the right, $f_2 > 0$) releases disinformation while a left-leaning party releases strong, high-quality information ($\sigma_1$ is large), the voters who consume both sources will actually shift further left than they would have without the disinformation. The logic is that the receiver, seeing a large value in $\xi_2$ but a moderate value in the strong $\xi_1$, infers that the noise in $\xi_2$ must be massive. Since the noises are correlated, they infer the noise in $\xi_1$ is also large, leading them to conclude the true signal $X$ must be small to satisfy the moderate $\xi_1$.

B. The Overshoot Effect

In a political spectrum $x_1 < x_2 < \dots < x_n$, if a candidate at $x_{n-1}$ uses disinformation to shift voters rightward:

• Moderate disinformation shifts voters toward $x_{n-1}$.
• Excessive disinformation increases the posterior probability of the extreme candidate $x_n$ more than $x_{n-1}$.
• This occurs because the posterior probability functions for intermediate candidates are bounded and unimodal, whereas the extremes can dominate if the signal is pushed far enough.

C. Continuous Time Dynamics

The results hold in continuous time (modeled via Brownian motion). The reversal condition $\sigma_2 < \rho \sigma_1$ remains valid, implying that even in a dynamic flow of information, a strong, high-quality signal can neutralize or reverse the impact of biased, lower-quality signals if the noise correlations are positive.

5. Significance and Implications

• Theoretical Insight: The paper provides a rigorous mathematical proof that disinformation is not always a zero-sum game where "more noise equals more confusion." Under specific structural conditions (positive noise correlation and high SNR in genuine sources), noise can be self-defeating.
• Policy and Strategy:
  • Counter-Strategy: The most effective defense against fake news is not merely debunking it (which requires the receiver to know the bias exists) but flooding the zone with high-quality, high-SNR information.
  • Fact-Checking Limitations: The authors argue that while fact-checkers are useful, their retrospective nature limits their impact. Instead, statistical knowledge of the distribution of disinformation (even without knowing specific instances) allows receivers to mitigate effects.
  • Risk of Radicalization: The "Overshoot Effect" warns political actors that aggressive disinformation campaigns can backfire by radicalizing the electorate toward the most extreme positions, bypassing the intended moderate target.
• Real-World Application: The paper cites the UK Conservative party's "Stop the Boat" campaign as a potential real-world example of the overshoot effect, where exaggerated messaging on migration inadvertently boosted support for the more extreme Reform UK party.

In conclusion, the paper demonstrates that the interaction between information sources is non-trivial. By leveraging communication theory, it reveals that the "weapon" of disinformation can be turned against the spreader if the information environment is structured correctly, offering a novel, mathematically grounded approach to protecting democratic processes.