The Interoception/Exteroception Boundary
Almost every page in this wiki uses the word interoceptive as though its extension were settled. It is not, and the sources disagree in a way that is easy to miss because none of them argues with the others.
The conventional answer, and why it fails
The default criterion is anatomical: the skin is the border, and interoception is the processing of signals generated below it. It has the great virtue of being checkable and the fatal defect of getting several important cases wrong in both directions.
- The vestibular system sits in the same sensory organ as audition, above any plausible internal border, and yet what it reports is the organism’s own balance. Anatomy files it outside; function puts it inside.
- Proprioception arises in muscles, tendons and connective tissue — anatomically internal — but reports body configuration in a way that has traditionally been grouped with somatosensation rather than with visceral sensing. See proprioception.
- Taste receptors sit at a surface that is topologically outside the body, sensing substances that have not yet entered it, yet gustatory afferents project to the NTS alongside unambiguously visceral ones.
- Chemoreception detached from the mouth entirely: functional bitter-taste receptors (TAS2Rs) on lung and vascular smooth muscle and on cortical neurons, possibly responding to bacterial ligands (Berntson & Khalsa 2021). Receptor type does not draw the line either.
- affective-touch: CT afferents lie in the skin, the paradigm exteroceptive organ, but share interoceptive fibre morphology and project to dorsal posterior insula rather than to S1/S2.
Craig’s 2002 widening was the first major redrawing — pulling temperature, pain, itch, sensual touch, hunger, thirst and air hunger inside on the grounds that they share the lamina I afferent route to the insula. That answer is pathway-based: interoceptive means carried by the homeostatic afferent system. It is the criterion most of this wiki has been implicitly using, and Berntson & Khalsa weakened it by showing that lamina I is one route among vagal, cranial, sacral, humoral and direct central chemosensing — some of which reach behaviour with no afferent fibre at all.
Three functional criteria, proposed within one year
All three appear in the Trends in Neurosciences 44(1) special issue, in different articles, with no cross-reference between them.
| source | criterion | what it makes the boundary about |
|---|---|---|
| Chen et al. 2021 | whether signals and responses represent, rather than originate from, the internal or external world | the content of the representation |
| Petzschner et al. 2021 | the inferred state the signal informs, not the channel it arrives on | the inferential role of the signal |
| Berntson & Khalsa 2021 | an act — ingestion is a decision to move a substance into the body, so the swallow may mark the transition | a behavioural transition point |
They agree on the important negative claim — the line is not drawn by anatomy or by receptor type — and they are not obviously inconsistent with each other. But they are answers to subtly different questions, and they come apart in practice. Chen et al.’s criterion makes the vestibular system interoceptive because balance is a fact about the organism. Petzschner et al.’s makes a signal interoceptive if it updates a model of bodily state, which could include a visual signal about one’s own body. Berntson & Khalsa’s locates the boundary at a moment in time rather than in a signal at all.
The pathway criterion’s obituary, stated as such (Desmedt et al. 2023)
The section above records that the pathway-based (“homeostatic afferent”) criterion was weakened. Desmedt, Luminet, Maurage & Corneille (2023) argue it should be abandoned as a definitional criterion altogether, and make it the central thesis of a full review rather than an aside. The argument is a modal one and it is worth stating in its own terms because it is the strongest version of this page’s negative claim.
The field, they observe, endorses both definitions at once — interoception as the processing of internal signals (phenomenon-based) and interoception as the activation of homeostatic pathways (physiological) — often in the same article. The physiological version equates interoception with slow, high-threshold afferent processing (C and Aδ fibres) and dissociates it from the fast, low-threshold Aβ processing assigned to somatosensation. Reviewing the cardiac, respiratory and gastro-intestinal systems in turn, they show each is served by both: alongside the homeostatic (baroreceptor, cardiac-wall, lamina I) routes, large-diameter Aβ somatosensory fibres also carry internal-state information — Pacinian masking degrades heartbeat detection (Knapp-Kline et al. 2021), the HEP is recordable in somatosensory cortex, and a bilateral-insula-lesion patient’s cardiac awareness vanished under chest-skin anaesthesia (Khalsa et al. 2009).
The conclusion: no physiological pathway is necessary and sufficient for the processing of internal states. Homeostatic pathways are not necessary (the same tissues are read by non-homeostatic ones) and not sufficient to mark the border (Aβ fibres carry external signals too). So physiology cannot draw the line — which is exactly this page’s thesis, arrived at from the fibre level rather than the case-by-case level, and pushed one step further into a positive recommendation: define interoception phenomenally, as processing of signals below the skin, excluding the classical exteroceptive senses to avoid overlap. Desmedt et al. are careful to leave standing the continuous physiological difference Carvalho & Damasio (2021) describe (interoception is proportionally more unmyelinated fibre); what they deny is a categorical physiological boundary. That is the same shape as the deflationary reading at the foot of this page — the boundary is real but graded, not a natural joint — now with a fibre-level mechanism attached.
A boundary attention can move within one brain
The criteria above locate the boundary in a signal — its pathway, its content, its inferential role. Farb, Segal & Anderson (2013) show that in the insula the boundary is not even fixed per signal: it moves with attention, within one person, across seconds. The same measured respiratory rate is tracked by the posterior insula (primary interoceptive cortex) when the breath is attended and by somatosensory cortex when a visual task is attended instead — one viscerosomatic signal, two cortical destinations, selected by where attention points. And along the insula’s own length the boundary is a gradient: posterior gyri tuned to interoception, anterior gyri better predicted by exteroceptive attention, so the anterior insula is by the authors’ own conclusion “not an area of pure body awareness.”
This does not adjudicate between the three criteria, but it constrains all of them. A definition that fixes a signal’s status by pathway or content has to accommodate the fact that the cortical representation of a fixed interoceptive signal is not fixed — it is an attentional achievement. It also puts an empirical floor under Petzschner et al.’s inferential-role criterion (a signal is interoceptive if it informs a bodily-state model): here the same signal informs an interoceptive or a somatosensory representation depending on task set. The neural counterpart is on insular-cortex; the deflation of the anterior insula’s body-awareness status is shared with Haruki & Ogawa (2023).
García-Cordero et al. (2017) show the electrophysiological face of the same point, and push it below the cortex-selection level to timing and frequency. Attending inward vs outward changes the frontal HEP within 200-500 ms (both interoceptive conditions more negative than a matched exteroceptive one), and — recorded intracranially from posterior insula, amygdala, somatosensory cortex and inferior frontal gyrus — flips the dominant band: interoception rides broadband high frequencies (35-110 Hz), exteroception low frequencies (1-35 Hz), in every one of those hubs. So the interoceptive/exteroceptive distinction, at the neural level, is at least partly a distinction between attentional modes with their own temporal and spectral signatures, not only between signals with fixed pathways. That cuts the same way Farb’s result does against the pathway and content criteria: whatever fixes a signal’s status, the brain’s handling of a fixed signal is reorganized by where attention points.
Why it matters rather than being a definitional quibble
Three concrete consequences, each already live on other pages.
It determines what the measures measure. If interoception is a natural kind unified by pathway, a cardiac accuracy score is a sample of a general capacity. If it is unified by what the signal represents, cardiac accuracy is a sample of one channel with no guarantee of generality — which is what Ferentzi et al. (2018) found empirically, reported on interoceptive-taxonomy and berntson-2021-neural-circuits: interoceptive sensitivity does not generalize across channels.
It determines the scope of the clinical claims. Chen et al. fold descending body regulation into interoception outright, which brings blood pressure, baroreflex and organ function inside as outcome measures. Whether a therapy “works through interoception” depends heavily on where this line is drawn — see weng-2021-interventions-of-interoception, where the organizing claim is exactly that and is nowhere measured.
It is quietly load-bearing for the constructionist arguments. If interoception is a party to multisensory-integration rather than a channel that terminates and then influences things (Quigley et al.), then asking whether a given percept is interoceptive or exteroceptive may be a badly-formed question about an assembled product. That is a stronger claim than any of the criteria above, and it dissolves the boundary rather than locating it.
The one case where the choice has now been cashed out empirically
Everything above is a disagreement about criteria. nociception is where the disagreement has started producing different readings of the same data, which makes it the most useful instance to track.
Craig’s widening put all pain inside; Ceunen, Vlaeyen & Van Diest (2016) put only visceral nociception inside; Critchley & Garfinkel (2017) record the split without resolving it. Wilzok et al. (2023) is the wiki’s first source to have to choose — they run pinprick pain as an experimental interoceptive channel alongside inspiratory loading — and they choose broad inclusion, on homeostatic grounds.
Two consequences, both live elsewhere in the wiki:
- Their cross-modal r = 0.57 is a within-interoception finding on the broad view and an interoception-to-somatosensation finding on the narrow one. See is-interoception-domain-general, where it now sits as a field position with that caveat attached.
- Existing evidence flips direction too. Ferentzi et al. (2018) and Crucianelli et al. (2022) both counted nociceptive measures among the channels failing to correlate with cardiac sensitivity, and the domain-generality debate counts those as interoceptive nulls. On the narrow definition they are partly nulls between interoception and something else — much weaker evidence. Neither paper states which definition it is using.
That is this page’s thesis made concrete: the boundary is not a philosophers’ problem downstream of the science. A definitional choice nobody argued for is doing work in the evidence tally of the wiki’s most consequential debate.
The boundary run from the exteroceptive side (Schwarzlose et al. 2023)
Every case above tests the boundary by asking whether some ambiguous signal (vestibular, taste, affective touch, pain) belongs inside. Schwarzlose et al. (2023) approach it from the unambiguously outside: sensory over-responsivity is distress at innocuous exteroceptive stimuli — sounds, textures, smells — and is transdiagnostically linked to childhood psychopathology and to altered salience/sensorimotor connectivity. That matters for this page not because SOR sits near the boundary (it does not) but because its neural signature lands on the same nodes — salience, ventral attention, amygdala — that the interoceptive over-responsivity literature implicates for over-attention to internal signals (anxiety-sensitivity, panic; see interoceptive-psychopathology).
So the wiki now holds a case where an interoceptive and an exteroceptive over-responsivity are described in the same vocabulary (over-precision on prediction error) and mapped to overlapping salience machinery. Read one way, that is evidence the boundary is real but the salience-weighting apparatus straddling it is shared — consistent with this page’s deflationary reading that the line is graded, not a natural joint. Read the other way, it is a warning: if the same machinery mis-weights signals on both sides, then a finding badged “interoceptive dysfunction” (e.g. Nord et al.’s transdiagnostic mid-insula) may be an instance of a domain-general sensory-precision failure that the interoceptive/exteroceptive label does not carve. See is-interoception-domain-general. The measure is thin (one parent-report item), so this is a framing parallel, not a settled mechanism.
Status
Open, and not currently contested — which is the odd part. Three criteria were proposed in one journal issue by authors who cite each other’s articles without engaging each other’s definitions. This page exists to keep them side by side rather than letting whichever was read most recently silently become the wiki’s working definition.