A World-Model Theory of Trait Anxiety
Year: 2026
Type: preprint
Abstract: Computational psychiatry has sought the mechanisms of anxiety in domain-general biases in learning and planning, and found small, inconsistent effects. We argue this is because trait anxiety is not a bias in how a person learns any task but a property of the person's world model, namely the beliefs they hold about threat, control, and the self in particular domains, and the strategies that bring those beliefs to bear on a situation. We formalize a world model as comprising causal knowledge, an intrinsic reward function, available actions, and internal strategies, and build a minimal generative model in which each has a concrete role in explaining common trait anxiety behaviors. Most importantly, the model inverts the usual place of generalization. Instead of asking whether a task's findings generalize to real-world conditions, it models how far a person's real-world beliefs generalize to the task, and treats the computational parameters of learning and planning that a task measures as contingent on that generalization. Simulations demonstrate how, within tasks similar-enough to the domains of core anxious beliefs, avoidance behavior and anxious thought can arise in qualitatively distinct ways by virtue of world model components. We review prior attempts to relate trait anxiety to computations of learning and planning using our model, and argue that, to advance the field, future studies should personalize cognitive tasks based on threat-related world model content.
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7213638693