A Cognitively Grounded Bayesian Framework for Misinformation Susceptibility

Researchers present Bounded Pragmatic Listener (BPL), a Bayesian framework that models how cognitive limitations affect susceptibility to misinformation. The framework incorporates three cognitively grounded constraints—working memory limits, information bottlenecks, and saliency-weighted sampling—to predict vulnerability to disinformation across benchmark datasets.

The research addresses a critical gap in understanding misinformation susceptibility by grounding computational models in cognitive science principles. Rather than treating information processing as unbounded rational analysis, BPL acknowledges that human cognition operates within strict resource constraints—limited working memory, compressed prior knowledge, and biased attention mechanisms that prioritize salient information. This approach bridges formal linguistic theory with bounded rationality literature, creating a framework that can predict not just whether someone accepts false information, but why certain cognitive architectures render individuals more vulnerable to specific types of misinformation.

The validation on LIAR and MultiFC benchmarks demonstrates practical applicability beyond theoretical contribution. The framework's ability to model differential vulnerability across mis-, dis-, and mal-information categories—distinctions often conflated in prior research—reflects sophisticated taxonomy that mirrors real-world information disorder. The depth-mismatch paradox finding suggests that certain cognitive limitations, while generally detrimental, can sometimes protect against manipulation when adversaries miscalibrate their deceptive strategies.

For technology platforms and content moderation teams, BPL offers a principled method to identify high-risk user segments and tailor interventions based on underlying cognitive constraints rather than surface-level behavioral patterns. The framework suggests that one-size-fits-all fact-checking approaches may fail because they don't account for how different cognitive limitations interact with specific misinformation types. This has implications for AI safety and human-AI interaction design, particularly as language models increasingly mediate information consumption.

Future work should extend BPL to real-world deployment contexts and explore interventions specifically designed around identified cognitive bounds rather than generic media literacy approaches.

• →BPL integrates bounded rationality principles with Bayesian pragmatics to model how cognitive constraints drive misinformation susceptibility
• →The framework identifies three key cognitive limitations: working memory depth, prior knowledge compression, and saliency-weighted attention
• →Differential vulnerability across mis-, dis-, and mal-information categories can be predicted and quantified using the model
• →Validation on benchmark datasets demonstrates the framework achieves competitive performance while providing interpretable predictions about user vulnerability
• →The depth-mismatch paradox suggests cognitive limitations sometimes inadvertently protect against sophisticated manipulation tactics