Debate at a glance
- Summary: Whether meditation and mindfulness training measurably improve performance on objective interoceptive-accuracy tasks (e.g., heartbeat detection), or instead primarily reshape attention habits, regulatory strategy, and subjective interoceptive constructs without a domain-general accuracy gain.
- Status: open
Positions in the field
- Sahib S. Khalsa and colleagues: Experienced meditators do not show superior heartbeat-detection accuracy relative to non-meditators in laboratory settings — contradicting practitioners’ self-perception of enhanced body awareness. (Khalsa et al. (2008), Interoceptive awareness in experienced meditators, Parkin et al. (2013), Exploring the relationship between mindfulness and cardiac perception)
- Norman A. S. Farb et al.: Mindfulness training more plausibly shifts interoceptive attention tendency and regulatory strategy (active toward perceptual inference) than domain-general accuracy; apparent benefits may be domain-specific (e.g., breath sensitivity) rather than global, and a multidimensional self-report measure may capture training effects (regulatory habits) that accuracy tasks miss. (Farb et al. (2015), Farb et al. (2010), Farb et al. (2013), Bornemann et al. (2014), Differential changes in self-reported aspects of interoceptive awareness through 3 months of contemplative training)
- Norman A. S. Farb et al. (Eusebio et al. 2022) — the randomized version, and it complicates the position above: Contemplative training did not raise self-reported interoception either — but it did produce behavioural transfer. In the wiki’s only RCT manipulating interoceptive emphasis while holding everything else constant, MAIA rose equally in both arms (group x time p = .443) and symptoms fell equally, while SART sustained-attention accuracy improved more under interoceptive instruction (p = .006, corrected). The Farb position above says accuracy tasks measure the wrong construct and multidimensional self-report captures what training actually changes. This trial ran that self-report measure under randomization and it did not discriminate. What discriminated was a behavioural outcome in a non-interoceptive domain, moderated by MAIA and by no other questionnaire — which relocates the evidence for training effects from self-report to transfer, without restoring accuracy. (Eusebio et al. (2022))
Connections
- Related debates: inferential?
- Related concepts: Interoceptive taxonomy (Farb et al. Box 1), Interoceptive sensitivity, Perceptual inference (interoceptive), body-focused contemplative training, Respiratory interoception
Surfaced explicitly in Farb et al. (2015) as a finding that “contradicts” a natural expectation: if meditators habitually direct attention to bodily sensation, one would expect superior heartbeat-detection accuracy — yet the laboratory evidence does not consistently show this (Khalsa et al. 2008; Parkin et al. 2013).
Why this matters
This is not a null debate — it has real stakes for interpreting every clinical and wellbeing claim in the wiki’s mindfulness literature. If “interoceptive training” doesn’t move the most commonly used objective metric (heartbeat detection), then claims about mindfulness improving interoceptive function must rest on some other construct within the Interoceptive taxonomy (Farb et al. Box 1) — attention tendency, regulation, coherence, sensibility — rather than accuracy per se.
Evidence complicating a simple “no effect” reading
Farb et al. marshal evidence that training effects are real but domain-specific or construct-specific rather than a general accuracy boost:
- Increased sensitivity in the specific domain trained (breath: Daubenmier et al. 2013).
- Reduced errors in body-scan-related somatic signal-detection tasks after even brief training (Mirams et al. 2013).
- Increased coherence between subjective and objective body-sensitivity measures (Fox et al. 2012) and between physiological and subjective states (Sze et al. 2010).
- Bornemann et al. (2014): the commonly assessed “observe” facet of mindfulness questionnaires changes far less with training than the regulatory aspects of interoceptive awareness — suggesting null accuracy findings may partly reflect measuring the wrong construct.
- Farb et al. (2010) supplies a neural version of the same point: MBSR completers, under a sad-film challenge, kept the interoceptive insula online where controls deactivated it, and insula recruitment tracked lower depression — a training-linked change in interoceptive engagement and regulation with no heartbeat task run at all. It cannot speak to accuracy directly, but it is a direct demonstration that something interoceptive changes with training even when the accuracy literature finds nothing — exactly the “right construct, wrong metric” pattern this position predicts. Caveat: cross-sectional (trained group scanned after, controls before), so it is an association, not a within-person training effect.
- [[farb-2013-training-interoceptive-cortex|Farb et al. (2013, SCAN)]] is the most direct neural instance of this position: MBSR graduates vs untrained controls attending to the breath in the scanner, where training altered where interoceptive attention is represented (greater dorsal anterior insula recruitment), deactivated DMPFC during interoceptive attention, and built a standing posterior↔middle-insula connectivity the untrained could only voluntarily invoke — with daily-practice dose predicting the primary posterior-insula signal (r=0.61). Again no accuracy task was run, and again cross-sectional. So it strengthens the claim that training reorganizes interoceptive attention and network architecture while leaving the accuracy question formally untouched — the position’s cleanest neural exhibit, and its clearest illustration of why the accuracy debate cannot be settled by studies that never measure accuracy.
Two additions from the intervention literature (Weng et al. 2021)
Neither settles the debate; both change what could settle it.
A measure that is neither a questionnaire nor a heartbeat. The impasse above is partly a measurement impasse: accuracy tasks may be the wrong construct, and self-report may be contaminated by exactly the belief that training instils. Weng et al. (2020) — the EMBODY framework (EMBODY framework (brain decoding of internal attention)), now held first-hand rather than through the 2021 review — trains participant-specific classifiers distinguishing interoception, mind-wandering and self-referential states, yielding a percentage of time in interoceptive focus during practice, with attentional location (breath vs feet) separately decodable. If the position on the right of this debate is correct — that training changes attention deployment rather than accuracy — this is the first instrument in the wiki that measures attention deployment directly. Two things temper it: it is a feasibility study (N = 16), and the meditation-period decode has no ground truth (Step 1 is validated on instructed trials, Step 2 is not). And decisively for this page: no published result yet uses it to test whether training changes attention deployment — the instrument that could settle the Farb-side claim has been demonstrated but never pointed at the claim. See Helen Y. Weng.
A clinical result that does not depend on resolving it. MABT trials report improved self-reported interoceptive awareness plus reduced substance use, craving and emotion dysregulation, and improved RSA, at 12 months. That the interoceptive outcome is self-report keeps it firmly on the disputed side of this debate — it is not evidence of accuracy gain and should not be cited as such. What it does show is that durable clinical benefit can be produced by a training designed around interoceptive awareness while this question remains open, which weakens the practical urgency of the debate without weakening its scientific force.
The randomized test, and the gap it leaves open
Everything above is cross-sectional or quasi-experimental: meditators compared with non-meditators, trained groups compared with waitlists. Eusebio et al. (2022) is the first source on this page with random assignment, an active comparator, and a longitudinal follow-up — and it manipulates interoceptive emphasis specifically while holding intervention format, duration, instructor training and social contact constant. That is the design this debate has needed.
Its result does not resolve the debate; it moves the impasse one step down.
| measured? | discriminated the arms? | |
|---|---|---|
| interoceptive accuracy | no — no accuracy task was administered | — |
| interoceptive sensibility (Multidimensional Assessment of Interoceptive Awareness (MAIA)) | yes | no (group × time p = .443) |
| symptoms, well-being, mindfulness (6 questionnaires) | yes | no |
| behavioural transfer (Sustained Attention to Response Task (SART)) | yes | yes (p = .006, corrected) |
What it costs the Farb position. That position’s escape from the null accuracy findings is that training changes regulation and attention tendency, and that multidimensional self-report — the MAIA specifically — captures what accuracy tasks miss. Under randomization, the MAIA missed it too. The escape route as stated is not available for this trial, and the Bornemann et al. (2014) evidence it leans on (regulatory subscales change more than the observe facet) was not reproduced here, because Eusebio et al. used total scores.
What it costs the Khalsa position. Nothing directly — no accuracy task was run — but it removes an inference that position has been able to rely on. If neither accuracy nor sensibility moves, one might conclude the training does nothing interoceptive. Eusebio et al. show something did change, differentially, and it showed up in behaviour: transfer to a task with no interoceptive content, moderated by interoceptive sensibility and by no other trait in the battery.
The gap. Nobody has run an interoceptive accuracy task before and after a training that was randomized against an active comparator matched on everything but interoceptive emphasis. Eusebio et al. built exactly that trial and then measured no accuracy. Adding a Heartbeat detection task or a respiratory threshold to that design is the single most decisive available study for this page, and it requires no new methodology.
A weak cross-sectional datum in the Khalsa direction
Ribera d’Alcalà et al. (2014) recorded self-reported years of mind-body practice (yoga, tai chi, meditation, Feldenkrais, Alexander technique) alongside heartbeat counting in 58 adults. More practice went with slightly lower accuracy (r = -.23, one-tailed p = .04), and a positive but non-significant trend appeared only after post hoc outlier removal. Treat it as noise-level: cross-sectional, self-reported dose, confounded (practice was concentrated in the older chronic pain group). Its only use here is directional: like Khalsa et al. (2008), it gives no sign that practitioners are more accurate. Its abstract reports the positive trend, not the negative correlation, so cite the results, not the abstract.
Open question
Whether future studies using the fuller Interoceptive taxonomy (Farb et al. Box 1) (rather than heartbeat-detection accuracy alone) will resolve this into “mindfulness changes regulation/attention, not accuracy” or instead find accuracy effects were simply underpowered/mismeasured remains unsettled. This connects to, but is distinct from, the inferential? debate about interoception’s basic processing architecture.