Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences 17(11), 565–573. DOI 10.1016/j.tics.2013.09.007.
Key takeaways
From the reading, NOTE-1000.
- Emotion as inference. Felt emotion is “the integrated content” of top-down predictions of interoceptive signals across levels, and “emotional content is determined by active inference on the likely internal and external causes of changes in the physiological condition of the body” (p. 567). This is offered as a generalisation of appraisal theories, from James and Lange to Schachter and Singer.
- Against bottom-up accounts. Damasio, Craig and Critchley ground the self in representations of the body too, but “none identify emotional states with top-down inference” (p. 569). Their hierarchy is “largely bottom-up”.
- Active interoceptive inference. Descending predictions act as set points for autonomic reflexes, as proprioceptive predictions do for motor reflexes in active-inference motor control. This requires lowering the precision of, or attention to, interoceptive prediction errors (p. 567, Figure 2).
- The anterior insula as comparator. The anterior insula is proposed as comparator, on indirect evidence: anticipation and prediction-error responses in fMRI, and false cardiac feedback that raises both its activity and the emotional salience of stimuli (pp. 567–569). The article says plainly that there is “not yet any direct confirmatory evidence” (p. 567).
- Body ownership. The rubber hand illusion is read as revision of a multisensory self-model to cut precision-weighted prediction error, and expecting correlated input is enough to induce it. People with lower heartbeat-detection sensitivity are more susceptible to it. A virtual hand flashing in time with the heartbeat is owned more, by questionnaire and by proprioceptive drift, unless finger movements are mapped to it. A whole-body version gave the same result (pp. 569–570, Figure 3).
- Psychopathology (Box 3). Depersonalization is proposed to come from imprecise interoceptive predictions, and Cotard’s delusion from explaining away persistent interoceptive error.
Standing in the record
Filed and read on 2026-10-19 at the owner’s request, to fill the record’s gap around the rubber hand illusion and bodily self-consciousness. It is the work that joins the record’s interoception cluster to body ownership. It was filed with Botvinick & Cohen (1998), Lenggenhager et al. (2007), Blanke & Metzinger (2009) and Apps & Tsakiris (2014), all four of which it cites or anticipates (Apps & Tsakiris as the 2013 online version).
Where it meets the record, set out in the NOTE:
- Seth’s later view (LIT-135, NOTE-169). The 2024 case for biological naturalism starts from interoceptive predictive processing. This is that view’s 2013 statement, about emotion and self and not yet about consciousness in general.
- Felt affect (THEORY-045). It takes the prediction side of the open question THEORY-045 leaves, prediction or map. It names Damasio’s account (LIT-388, which it cites) as bottom-up, and it cites the false-feedback evidence THEORY-045’s promotion test asks for, at second hand. Barrett’s later account (LIT-542, NOTE-434) cites it.
- The self-model (LIT-450, NOTE-372, THEORY-064). It gives Metzinger’s body-ownership claim (NOTE-372, C6) a mechanism and an interoceptive route. It also gives a self-model that includes the body’s interior, which Metzinger’s article names only as an unconscious layer.
- Active inference (LIT-1181, NOTE-960). It carries Friston’s account of motor control to autonomic control.
Filed alongside, in the 2026-10-19 batch on the rubber hand illusion and bodily self-consciousness: Botvinick & Cohen 1998, the rubber hand illusion (LIT-1277); Lenggenhager et al. 2007, the full-body illusion (LIT-1279); Blanke & Metzinger 2009, minimal phenomenal selfhood (LIT-1274); Apps & Tsakiris 2014, the free-energy self (LIT-1262).
Filed from this work on 2026-10-20: THEORY-767, on interoceptive signals in body ownership.