Strategic communication of narratives: An experiment

TEXT START: We investigate the strategic communication of narratives under model uncertainty.

The Dissection

The paper models narrative communication as a strategic game: an informed sender interprets uncertain data for a receiver, while the receiver discounts the message according to the sender’s incentives. Its experiment finds that stronger sender bias produces greater exaggeration and stronger receiver correction. Narratives are only slightly persuasive when the sender’s bias is low.

The text is really measuring whether humans can detect and compensate for motivated interpretation in a controlled environment. That is a useful micro-result. It is not an account of the industrialization of communication, synthetic media, automated persuasion, or the collapse of human economic participation.

The Core Fallacy

The central error, under Discontinuity Thesis mechanics, is confusing local epistemic adaptation with systemic resilience.

A receiver’s ability to discount a biased narrator does not constrain AI’s ability to generate narratives at negligible marginal cost, personalize them, multiply them across channels, test them continuously, and coordinate them at scale. The experiment assumes a relatively legible sender, a bounded interaction, identifiable incentives, and enough attention for the receiver to perform correction. Those conditions are precisely what mass synthetic communication erodes.

The paper studies whether people can resist a narrator. The DT question is whether human institutions can preserve a stable human-only information environment once narrators become automated, abundant, adaptive, and strategically networked. Under P1 and P2, they cannot. The experiment therefore does not weaken the thesis. It measures antibodies while the pathogen is being industrialized.

Hidden Assumptions

• Human senders and receivers remain the relevant strategic actors.
• Sender bias is stable, observable, and inferable.
• The data-generating process remains sufficiently accessible for correction.
• Receivers have the time, attention, and competence to audit narratives.
• Communication occurs in bounded interactions rather than high-volume, coordinated networks.
• Synthetic identity, consensus manufacturing, and automated repetition are absent or negligible.
• Persuasion is measured through immediate recommendations rather than cumulative belief formation.
• Receiver correction remains cognitively available after AI automates both advocacy and counter-advocacy.
• The social value of communication can be separated from the ownership and control of the systems producing it.

That last omission matters. Even perfect narrative skepticism does nothing to restore productive participation once AI severs the employment–wage–consumption circuit. The paper does not touch P3.

Social Function

Primary classification: partial truth.

Secondary classification: prestige signaling when its findings are inflated into evidence that rational human correction can govern the emerging information order.

The paper establishes that people are not perfectly credulous and may account for strategic incentives. That is real. But its institutional comfort is misleading: it frames the problem as better inference inside a manageable communication game, not as the displacement of human communicators by scalable machine systems. It offers cognitive adjustment where the structural problem is ownership, coordination, and control.

The Verdict

This is a competent laboratory study of motivated interpretation, not a defense of the existing economic order. Its result is narrow: humans can sometimes recognize biased narratives and discount them. Under the DT framework, that capability is a temporary defensive adaptation, not a moat. AI can automate the sender, the counter-sender, the audience segmentation, and eventually the verification layer itself. The paper records how humans respond to strategic narratives just before strategic narration ceases to be a predominantly human activity.