When Rituals Fail: Rationalization, Bayesianism, and Predictive Processing (onlinelibrary.wiley.com)

This paper argues that the human tendency to ritualize and to rationalize failures can be fruitfully analyzed through Bayesian and predictive‑processing frameworks. The author frames ritual behaviors as strategies for minimizing prediction error in hierarchical generative models: rituals instantiate strong priors and low‑variance action policies that stabilize perception and social expectations. When rituals “fail” — i.e., when outcomes contradict entrenched expectations — people tend to engage in post‑hoc rationalization rather than full Bayesian belief revision. The paper connects that phenomenon to mechanisms in predictive processing such as precision weighting of prediction errors, prior rigidity, and active inference: low precision assigned to disconfirming signals or high policy priors can preserve belief despite contrary evidence, producing systematic rationalization. For the AI/ML community the work highlights technical implications: modeling human-like persistence and biased updating requires accounting for asymmetric precision and model misspecification, not just normative Bayesian updates. It suggests incorporating mechanisms for prior rigidity, selective precision modulation, and action‑based hypothesis testing into cognitive architectures and generative models to better simulate social learning, confirmation bias, and belief resilience. Practically, these ideas inform robust inference under out‑of‑distribution observations, human–AI interaction design that anticipates rationalization, and interpretability frameworks that separate model evidence from policy‑driven stabilization.