🧬 biology

Neurocomputational mechanisms of sharing clarifying information about ambiguous news

By combining a coordination game with computational modeling and fMRI, this study reveals that the decision to share clarifying information about ambiguous news relies on a hierarchical Bayesian process where the brain integrates confidence-based beliefs about information value with social-cue-inferred beliefs about recipients' preferences, mediated by distinct neural circuits in the vmPFC, ventral striatum, TPJ, and lateral frontopolar cortex.

Original authors: Jean-Claude Dreher, Valentin Guigon, Julien Benistant, Marie Villeval

Published 2026-09-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jean-Claude Dreher, Valentin Guigon, Julien Benistant, Marie Villeval

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

In the digital age, the spread of false or misleading information is often blamed on people who cannot tell the truth from a lie. Yet, a deeper problem often lies in what happens after a person forms an opinion. Even when someone is unsure about a piece of news, they might choose not to share a clarifying fact if they believe the person receiving it would not want to hear it. This decision is not just about the news itself; it is a social calculation. The sender must guess what the receiver knows, what they believe, and whether they are open to having their uncertainty reduced. This process, known as inferring another person's informational preferences, is a fundamental part of how knowledge moves through a society. If people misjudge what others want to know, valuable corrections get withheld, and confusion persists. Understanding the brain mechanisms behind this guesswork is crucial for figuring out why misinformation spreads and how to stop it.

A team of researchers in France set out to map the specific neural machinery that allows people to make these guesses. They designed an experiment where participants, acting as senders, had to decide whether to send extra information about an ambiguous news story to a receiver. The news items were deliberately vague, neither clearly true nor clearly false, forcing the senders to rely on their own judgment and their guess about the receiver's mind. In some trials, the senders had no idea who the receiver was or what they thought. In other trials, they were given a subtle clue: a visual indicator of how socially close the receiver was to an organization associated with the news topic. For example, if the news was about climate change, the sender might see how close the receiver's values were to a group that disputes human involvement in climate change. The senders were rewarded only if their decision to send or withhold information matched the receiver's actual desire to receive it.

The researchers found that people use two different mental strategies depending on the information available. When they had no clue about the receiver, they relied on their own confidence. If a sender was very unsure about whether a news item was true, they assumed the receiver was also unsure and would appreciate more information. If the sender felt certain, they assumed the receiver did not need help and withheld the extra details. However, as soon as a social clue was introduced, the senders shifted their strategy. They stopped relying solely on their own feelings of certainty and started using the social cue to estimate the receiver's specific preferences. This shift was not just a change in behavior; it was a complete reorganization of how they weighed evidence. The study showed that people are remarkably flexible, able to switch from projecting their own mind onto others to using external social signals to build a more accurate picture of what another person wants.

To understand how the brain manages this switch, the researchers used functional magnetic resonance imaging to watch the participants' brains in real time. They discovered that different parts of the brain handle different parts of this social calculation. When senders relied on their own confidence to guess what the receiver wanted, activity increased in the ventromedial prefrontal cortex and the ventral striatum. These are regions known for evaluating value and processing personal certainty. Essentially, the brain was using its own sense of "I am unsure" as a proxy for "they are unsure." When the social clue became available, a different region, the temporoparietal junction, took over. This area is famous for its role in understanding other people's thoughts and intentions. It encoded the specific information about the receiver's social distance, allowing the sender to update their guess based on the new evidence.

The final step of the decision-making process happened in the lateral frontopolar cortex, a region at the very front of the brain. This area acted as a conductor, integrating the personal confidence signal and the social clue into a single, unified estimate of what the receiver wanted. It was here that the brain combined the internal feeling of uncertainty with the external social data to make the final call on whether to send the information. The study suggests that this integration is a distinct process, separate from simply judging the news or evaluating one's own knowledge. It is a specific mechanism for coordinating with another person's mind.

The findings offer a clear picture of how the brain navigates the complex social landscape of information sharing. It shows that when we decide to share or hide a piece of news, we are not just reacting to the content. We are running a sophisticated, multi-step calculation that weighs our own uncertainty against our best guess of another person's needs. When we lack information about the other person, we default to using our own feelings as a guide. But when we have a social cue, even a small one, our brains rapidly reweight that information, shifting from self-projection to a more accurate reading of the other person. This ability to infer what others want to know is a critical, yet often invisible, part of how truth travels through our networks. If this inference process is disrupted, or if the social cues are too weak to trigger it, clarifying information may never reach the people who need it most, allowing ambiguity to fester and misinformation to spread.

Technical Summary: Neurocomputational Mechanisms of Sharing Clarifying Information about Ambiguous News

Problem Statement
Misinformation propagation is driven not only by errors in judging the veracity of ambiguous news but also by the withholding of clarifying information when recipients would benefit from receiving it. While existing research characterizes the neural mechanisms of information seeking and self-motivated sharing, the upstream neurocomputational processes by which individuals infer a recipient's informational preferences—particularly when those preferences are uncertain—remain unknown. Specifically, it is unclear how individuals integrate their own uncertainty about a claim (first-order beliefs) with inferences about a recipient's stance (second-order beliefs) to guide the selective transmission of clarifying information.

Methodology
The study employed a novel coordination game, the "Sending News game," combined with computational modeling and model-based functional magnetic resonance imaging (fMRI).

Experimental Design: Thirty-two participants (Senders) performed tasks inside an fMRI scanner. On each trial, Senders evaluated the veracity of ambiguous news items, reported their confidence, and decided whether to transmit additional clarifying information to a Receiver. Crucially, Senders never saw the content of the additional information; their reward depended solely on matching the Receiver's revealed preference to receive or avoid that information.
Conditions: Two conditions were tested:
1. Control: Senders had no information about the Receiver, mimicking anonymous interactions.
2. Cue: Senders received a visual cue indicating the social distance between the Receiver and an organization associated with the news domain, mimicking partial social knowledge available on digital platforms.
Behavioral Data: Receivers (a separate group of 20 participants) had previously completed a "Receiving News game" where they indicated their willingness to receive clarifying information. Sender decisions were matched against this fixed dataset.
Computational Modeling: The authors compared heuristic, reinforcement-learning, utility-based, and hierarchical Bayesian models. The winning model, BayesRL, was a hierarchical Bayesian model that integrated four belief components: confidence-derived veracity evidence, social-distance evidence, feedback-based learning, and bias.
Neuroimaging: fMRI data were analyzed using General Linear Models (GLM) with parametric modulators derived from the latent belief modes of the BayesRL model ( $\theta_{truth}$ , $\theta_{distance}$ , $\theta_{learning}$ , and the integrated posterior $\theta_{action}$ ).

Key Results

Behavioral Findings:
- In the absence of social cues (Control), Senders used their own confidence as a proxy for the Receiver's preference: lower confidence increased the probability of sending clarifying information.
- In the presence of social cues (Cue), Senders shifted reliance toward Receiver-specific second-order beliefs. Sending probability decreased as the social distance between the Receiver and the relevant organization increased.
- Senders adapted their decisions over time based on feedback, with a steeper decline in sending for items judged false in the Cue condition.
Computational Findings:
- The BayesRL model significantly outperformed alternative models (heuristic, RL, utility-based) in explaining participant behavior (Protected Exceedance Probability = .99).
- Model parameters revealed a reweighting of evidence: in the Control condition, decisions were driven by confidence-derived veracity beliefs and priors; in the Cue condition, the weight of the social-distance component increased markedly (nearly threefold), while the weight of veracity and learning components decreased.
Neural Findings:
- First-Order Beliefs: Activity in the ventromedial prefrontal cortex (vmPFC) and ventral striatum tracked the confidence-derived veracity component ( $\theta_{truth}$ ). These regions encoded the subjective value of information derived from the Sender's own uncertainty.
- Second-Order Beliefs: Activity in the right temporoparietal junction (TPJ) and superior temporal sulcus tracked the social-distance component ( $\theta_{distance}$ ), encoding inferences about the Receiver's preferences based on social cues.
- Integrated Decision Signal: Activity in the lateral frontopolar cortex (LFPC) tracked the integrated posterior estimate ( $\theta_{action}$ ), which combined first-order, second-order, and learning components into a unified estimate of the Receiver's preference that directly governed the transmission decision.

Significance and Claims
The paper claims to identify a distinct neurocomputational architecture for "epistemic coordination"—the process of inferring whether another person wants clarifying information. The authors argue that this process is dissociable from both self-directed information seeking and self-related sharing motives (e.g., social approval).

Mechanism of Inference: The study demonstrates that when social information is absent, individuals rationally project their own epistemic state (confidence) onto others. When social cues are available, the brain shifts to a more specific second-order inference mechanism.
Neural Dissociation: The findings establish a hierarchical neural architecture where valuation regions (vmPFC/striatum) encode private uncertainty, mentalizing regions (TPJ) encode social inferences, and the lateral frontopolar cortex (LFPC) acts as an integrator, synthesizing these heterogeneous signals to guide action.
Implications for Misinformation: The authors suggest that on social media platforms, where incentives often prioritize self-value (likes, virality) over accuracy and social cues are sparse, the self-value signals in the vmPFC and striatum may override other-directed epistemic inference. This mechanism may explain why clarifying information is often withheld or why ambiguous content spreads indiscriminately. The study posits that understanding this specific inference process offers a potential target for interventions to improve information flow, distinct from strategies targeting mere veracity evaluation.

The authors maintain a modest scope, noting that their design isolates the inference of preference from content evaluation and uses neutral, non-partisan stimuli. They acknowledge that the findings may differ in naturalistic, reciprocal, or politically charged contexts where content framing and emotional engagement play larger roles.

More like this