A game of information

This paper analyzes a game where two players strategically select signal-to-noise ratios to influence a receiver's assessment, demonstrating that the problem reduces to an infinite game on a square with a complete equilibrium solution and proposing three generalizations.

The Gist

The Big Picture: A Battle of Noisy Megaphones

Imagine two people, Player A and Player B, standing on opposite sides of a town square. They are both trying to convince the crowd (the "Receiver") that a specific binary fact is true. Let's say the fact is: "Is the sky blue?" (Yes or No).

• Player A wants the crowd to believe "Yes."
• Player B wants the crowd to believe "No."

However, neither player can speak perfectly. Their voices are carried by megaphones that are filled with static (noise).

• The Signal is the actual message they want to send.
• The Noise is the static that distorts the message.
• The Signal-to-Noise Ratio is how loud and clear the megaphone is. If you turn the volume up (higher ratio), the message is clearer. If you turn it down, the static drowns it out.

Both players have a limited budget. They can't just turn their megaphones to "Maximum Volume" forever; they have to choose a specific volume level (let's call it $\sigma$) to stay within their budget.

The crowd is smart. They listen to both megaphones at the same time, combine the sounds, and use logic (math called "Bayes' rule") to decide what they believe. The goal of the game is for each player to choose the perfect volume level to sway the crowd, without knowing what volume the other player has chosen.

The Core Discovery: It's a Simple Math Game

The paper shows that even though this sounds complicated, it can be boiled down to a very simple math game played on a square grid (from 0 to 1).

The outcome depends heavily on how the "static" in the two megaphones relates to each other. This relationship is called Correlation ($\rho$).

• Negative Correlation: When one megaphone has static, the other is quiet.
• Positive Correlation: When one megaphone is loud with static, the other is also loud with static.

The author found three distinct strategies (equilibria) depending on this correlation:

1. The "Silence and Shout" Strategy (Negative Correlation):
   If the noise is negatively correlated, the smartest move is for Player A to turn their megaphone completely off (Volume = 0) and let Player B shout as loud as possible (Volume = 1). Surprisingly, having no signal from one side helps the other side's signal stand out more clearly against the combined noise.

2. The "Balanced Shout" Strategy (Low Positive Correlation):
   If the noise is slightly positive, Player A should shout at a specific, moderate volume (related to the correlation), while Player B shouts at maximum volume.

3. The "Coin Flip" Strategy (High Positive Correlation):
   If the noise is very similar (highly correlated), the game gets tricky. Neither player can pick just one volume. Instead, Player B must play a mixed strategy. This means Player B acts like a coin flipper: sometimes they shout at maximum volume, and sometimes they stay silent, based on a specific probability. Player A, meanwhile, picks a fixed volume that makes them indifferent to Player B's coin flips.

The Zero-Sum Nature:
The paper notes this is a "zero-sum game." If Player A gains a 10% increase in the crowd believing them, Player B loses exactly 10%. There is no middle ground where both win; one's gain is the other's loss.

What Happens When Someone Cheats? (Disinformation)

The paper also explores a scenario where Player B decides to cheat. Instead of just adding noise, they add Disinformation.

• Noise: Random static (like a bad connection).
• Disinformation: A deliberate lie or a "drift" in the signal intended to mislead the crowd.

Imagine Player B isn't just shouting through a static-filled megaphone; they are whispering a lie that slowly shifts the crowd's opinion over time.

The paper finds a fascinating twist here: The cheater can accidentally hurt themselves.
Because the crowd combines the two signals, if Player A increases their volume (signal strength), they can actually cancel out or even reverse the effect of Player B's lie. It's like if someone tries to push a heavy cart in the wrong direction, but you push it from the other side so hard that the cart actually moves the right way anyway.

However, because the cheater's "lie" (the drift) breaks the mathematical symmetry, the paper admits there is no simple, clean formula to solve this version of the game. It becomes a messy puzzle that requires computer simulations to solve.

Other Ideas Mentioned

The author briefly mentions two ways to expand this game, though they don't solve them fully:

1. More Players: What if there are three candidates instead of two? The math gets much more complex, involving a web of correlations between all three.
2. Changing Volume Over Time: Instead of picking one volume for the whole day, what if players can change their volume minute-by-minute? This turns the game into a "functional game," where they are optimizing a whole curve of volume changes rather than a single number.

Summary

In short, the paper argues that when two sides try to influence public opinion using noisy information, the best strategy depends entirely on how "connected" their noise is. Sometimes, the best move is to stay silent; sometimes, you must shout; and sometimes, you must gamble with a coin flip. If one side tries to cheat with lies, the other side can sometimes use a louder, clearer voice to neutralize the deception.

Technical Summary

Technical Summary: A Game of Information

Problem Statement
The paper addresses a strategic interaction between two players, A and B, who disseminate information regarding a binary topic of interest (represented by a random variable $X \in \{0, 1\}$) to a rational receiver (the public). The information transmitted is obscured by noise. Both players preselect their information flow rates (signal-to-noise ratios), denoted $\sigma_A$ and $\sigma_B$, subject to budget constraints ($\sigma_A, \sigma_B \leq 1$). The receiver combines these two noisy sources and updates their belief about $X$ using Bayes' rule (specifically, the Wonham filter).

The players have opposing objectives: Player A aims to maximize the probability that the receiver's posterior belief $\pi_T$ exceeds a threshold $K$ (swaying the public in one direction), while Player B aims to minimize this probability. The core problem is to determine the optimal strategies for $\sigma_A$ and $\sigma_B$ given that the players do not know the other's choice, effectively formulating a zero-sum game. The paper initially assumes "fair" information (noise only) before extending the model to include disinformation.

Methodology
The author models the information transmission using communication theory principles, where the signal is a fixed binary variable and noise is modeled as additive Gaussian processes (standard Brownian motions) with a correlation coefficient $\rho \in [-1, 1]$.

1. Information Fusion: The paper derives the effective signal-to-noise ratio ($\sigma$) resulting from the combination of two correlated information sources. The combined process is shown to be equivalent to a single source with an effective flow rate $\sigma$ and effective noise, derived via linear combination of the individual sources.
2. Bayesian Updating: The posterior probability $\pi_t$ is expressed as a function of the combined signal. The probability of the receiver making a specific choice (e.g., $\pi_T > K$) is calculated using the cumulative normal distribution function, dependent on the effective $\sigma$.
3. Game Theoretic Reduction: The problem is reduced to an infinite game on the unit square $[0, 1] \times [0, 1]$. The analysis focuses on the convexity of the effective flow rate $\sigma$ with respect to the players' strategies. The paper investigates the existence and form of Nash equilibria (pure or mixed) based on the correlation parameter $\rho$ and the prior probability $p$.
4. Extensions: The methodology is extended to:
   • Disinformation: Introducing a deterministic drift term ($\mu$) in one player's signal, representing intentional bias.
   • Multi-player Games: Generalizing the fusion formula to three players using inverse correlation matrices.
   • Time-Dependent Strategies: Formulating the problem where players choose flow rates $\sigma_t$ over a continuous time interval $[0, T]$, reducing the problem to a functional game.

Key Results
The paper provides a complete analytical solution for the time-independent, fair-information game:

• Nash Equilibrium Structure: The optimal strategies depend critically on the noise correlation $\rho$ (assuming $p > 1/2$):
  • Case 1 ($-1 < \rho < 0$): A pure strategy equilibrium exists where Player A chooses $\sigma_A^* = 0$ and Player B chooses $\sigma_B^* = 1$.
  • Case 2 ($0 \leq \rho < 1/\sqrt{2}$): A pure strategy equilibrium exists where Player A chooses $\sigma_A^* = \rho$ and Player B chooses $\sigma_B^* = 1$.
  • Case 3 ($1/\sqrt{2} \leq \rho < 1$): A mixed strategy equilibrium emerges. Player A chooses a fixed $\sigma_A^* = 1/(2\rho)$, while Player B randomizes between $\sigma_B = 0$ and $\sigma_B = 1$ with specific probabilities derived from the indifference condition.
• Zero-Sum Nature: The game is confirmed to be zero-sum regarding the change in the probability $P(\pi_T > 1/2)$, as the gain for one player exactly matches the loss for the other due to the antisymmetry of the probability function about $p=1/2$.
• Disinformation Impact: When Player B introduces disinformation (drift $\mu$), the symmetry of the game is broken. The paper notes that if $\rho > 0$, Player A can counteract the disinformation by increasing their own flow rate, potentially reversing the intended effect. However, a closed-form solution for the disinformation game is not provided; it requires numerical analysis.
• Generalizations:
  • For three players, the effective flow rate remains a quadratic form of control variables involving the inverse correlation matrix, though the equilibrium solution is not derived in closed form.
  • For time-dependent strategies, the problem reduces to optimizing the integral of the squared effective flow rate $\int_0^T \sigma_s^2 ds$, transforming the game into a functional optimization problem.

Significance and Claims
The author claims that the primary contribution of this work is demonstrating that a complex problem involving information dissemination, Bayesian inference, and strategic interaction can be translated into a standard infinite game on the unit square. This reduction allows for a complete equilibrium solution to be derived in closed form for the time-independent case.

The paper positions the "game of information" as a framework applicable to various real-world scenarios, including electoral competitions, marketing strategies, and climate policy debates. It specifically highlights the distinction between noise (unintentional errors) and disinformation (intentional bias), noting that while the former is handled by the derived equilibrium, the latter complicates the strategic landscape by breaking symmetry. The author concludes by identifying the derivation of closed-form solutions for multi-player games and time-dependent functional games as open challenges for future research.