Pain as a predictive process in the brain: how the Bayesian brain generates, amplifies, and relieves pain

Pain is not a passive readout of nociceptor traffic; it reflects the brain's inference about bodily threat under uncertainty. Building on predictive processing, active inference, and the free energy principle, we frame acute, chronic, and phantom pain as regimes of hierarchical inference in which memory, affect, and context shift the relative influence of prior beliefs vs incoming nociceptive evidence. Prior work links expectations to pain, but translating these ideas into a clinically usable account that links expectation mechanisms to measurable proxies of belief updating and to principled routes for personalized intervention remains challenging. We propose a generative model that links inferred threat to action policies, so that pain intensity and pain behavior can be studied within the same inferential loop. Within this framework, chronic pain can persist when threat beliefs remain overly influential despite corrective evidence. Placebo/nocebo effects illustrate how shifting expectations can amplify or relieve pain. Phantom pain illustrates how stable body-related beliefs can sustain pain-like perception even with minimal or absent afferent input. We outline candidate measurements and falsifiable predictions by linking expectation- and precision-related effects to a cingulate-insula-centered hub and associated descending control circuitry (periaqueductal gray to spinal dorsal horn). This account suggests principles for selecting and sequencing treatments to target belief content (expectations), evidence (corrective sampling), and gain control (context and neuromodulation), for matching mechanisms to patient-specific phenotypes, and for stratifying patients by dominant mechanism (overly threat-biased beliefs, reduced corrective sampling, or biased precision/gain control) to enable more testable trials and clearer interpretation of treatment response.