Interoceptive trait prediction error (ITPE)

Introduced in Seth & Friston (2016) to operationalize a stable individual-differences version of interoceptive prediction error, distinct from the trial-by-trial prediction errors posited by the core interoceptive-inference model. Computed as the gap between:

• objective interoceptive accuracy — performance on a standard heartbeat-detection-task, and
• subjective interoceptive sensibility — self-reported confidence/questionnaire measures of one’s own interoceptive ability.

Autism findings

A study cited in Seth & Friston (2016) found autistic individuals show both (i) reduced objective interoceptive accuracy and (ii) increased subjective interoceptive sensibility relative to controls — the combination yielding a larger ITPE than in controls. Across both autistic and control participants, ITPE magnitude correlated with self-reported anxiety, echoing Paulus & Stein’s earlier (non-Bayesian) proposal linking anxiety to interoceptive prediction error.

Complications

• Autism frequently co-occurs with alexithymia (difficulty identifying/describing one’s own emotions); one cited study found atypical interoception tracked alexithymia rather than autism per se, without directly testing ITPE — leaving open how cleanly ITPE dissociates from alexithymia.
• Findings on the direction of subjective interoceptive-sensibility change in autism are inconsistent across studies (at least one finds decreased subjective body awareness), which the paper attributes to autism’s broad heterogeneity resisting single-process explanations.

Relation to the wider taxonomy

ITPE is conceptually adjacent to — but not identical with — the accuracy/sensibility/awareness distinctions already tracked in interoceptive-taxonomy (Farb et al.’s seven-construct breakdown) and flagged as a terminology-in-flux area on interoceptive-sensitivity. Where Garfinkel et al. treat accuracy and sensibility as separate axes to be measured and reported independently, ITPE explicitly combines them into a single discrepancy score and gives that discrepancy theoretical status as a predictive-coding quantity (a stable, trait-level “prediction error”), rather than treating over/under-confidence as merely an interesting correlational pattern.