5D Library › Interoceptive Decoupling, Salience-Network Switching and Autonomic Visceral Restoration · 中文

Chapter 1. Interoception and Precision: The First Link Runs Backwards

The interoceptive pathways

“Interoception” means the perception of the body’s internal state: the heartbeat, the breath, gut motility, bladder filling, vascular pulsation, the chemical environment of tissue.

Evidence

Anatomy and physiology (review). Small-diameter afferents from viscera and tissue (Aδ and C fibres) enter through the dorsal horn, synapse in lamina I, ascend in the spinothalamic tract to the parvicellular part of the ventromedial thalamic nucleus, and project to posterior insular cortex; Craig described this pathway as a primate-specific system supporting a “cortical representation of bodily state” (Craig 2002, 2009). A separate vagal afferent route ascends via the nucleus of the solitary tract and the parabrachial nucleus. The insula shows a posterior-to-anterior gradient of re-representation, and anterior insula together with anterior cingulate participates in integrating emotion and bodily state (Craig 2009; Critchley and Harrison 2013). Critchley and colleagues found that individual differences in heartbeat awareness correlate with grey-matter volume in right anterior insula and with its response to the heartbeat (Critchley et al. 2004).

What has to be kept clear is that the above is evidence about pathways and correlations. It establishes that bodily state has a cortical representation; it does not establish that regulating the representation regulates the body.

Note

Do not treat the insula as a switch. The insula participates in almost every task involving bodily state, risk, pain or emotion, as Uddin’s review sets out (Uddin 2015). A region that activates in many tasks cannot be used to establish one particular causal story; this is the standard problem of reverse inference, discussed in Volume 33, Chapter 2.

Writing interoception as inference

For about fifteen years interoception has increasingly been written as inference: the brain does not read bodily state directly but maintains a generative model of it and updates that model with prediction errors. Barrett and Simmons call this interoceptive prediction (Barrett and Simmons 2015), Seth and Friston place it in the active-inference framework (Seth and Friston 2016), and Petzschner and colleagues apply it under the name “computational psychosomatics” (Petzschner et al. 2017). Its mathematical basis is predictive coding (Rao and Ballard 1999) and the free-energy principle (Friston 2010); in that framework attention is written as the modulation of precision (Feldman and Friston 2010). Volume 15 is devoted to this framework; here only its simplest form is needed.

The simplest form is the combination of two Gaussian sources. Let the bodily state be xx, the prior 𝒩(μ0,π0−1)\mathcal{N}(\mu_0, \pi_0^{-1}), and the interoceptive sample yy with precision πy\pi_y (precision being the reciprocal of variance). The posterior is

π1=π0+πy,μ1=π0μ0+πyyπ0+πy.\pi_1 = \pi_0 + \pi_y, \qquad \mu_1 = \frac{\pi_0 \mu_0 + \pi_y\, y}{\pi_0 + \pi_y}.

These two expressions are all the mathematics in this chapter, and the first result reads straight off them. The original claim asks for a reduction in the precision of the interoceptive channel, that is, a smaller πy\pi_y. Look at μ1\mu_1: reducing πy\pi_y pushes μ1\mu_1 toward μ0\mu_0. The posterior moves toward the prior. If the prior is benign (μ0\mu_0 near baseline) the posterior estimate falls and tension eases; if the prior is itself alarmed (μ0\mu_0 greater than the current bodily signal) the posterior estimate rises.

Calculation

# Chapter 1. Interoceptive inference as precision-weighted cue combination.
# Two Gaussian sources, a prior (expected bodily state) and an interoceptive
# signal, combine into a posterior whose precision is the sum of the two
# precisions:
#   pi_post = pi_prior + pi_sense,  mu_post = (pi_prior mu_prior + pi_sense y)
#   / pi_post.
# "Turning down the precision of the interoceptive channel" therefore moves
# the posterior TOWARD THE PRIOR and shrinks posterior variance less; it does
# not by itself make the posterior benign. This is the dark-room objection in
# one line: if the prior is itself alarming, lowering sensory precision makes
# the estimate MORE alarming, not less. All numbers are dimensionless
# illustrative units.
import math

def combine(mu_prior, pi_prior, y, pi_sense):
    pi_post = pi_prior + pi_sense
    mu_post = (pi_prior * mu_prior + pi_sense * y) / pi_post
    return mu_post, pi_post

def kl_gauss(m1, p1, m0, p0):
    """KL( N(m1,1/p1) || N(m0,1/p0) ) in nats."""
    return 0.5 * (math.log(p1 / p0) + p0 / p1 + p0 * (m1 - m0) ** 2 - 1.0)

if __name__ == "__main__":
    # an interoceptive signal 2 units from the benign set point
    Y = 2.0
    print("A. benign prior (mu_prior = 0): lowering sensory precision helps")
    print("   pi_sense   mu_post   sd_post   |mu_post| vs signal")
    for pi_s in (4.0, 2.0, 1.0, 0.5, 0.25):
        m, p = combine(0.0, 1.0, Y, pi_s)
        print("   {:8.2f} {:9.3f} {:9.3f}   {:5.0f}% of the signal"
              .format(pi_s, m, 1 / math.sqrt(p), 100 * m / Y))

    print(
        "\nB. alarmed prior (mu_prior = 3): the SAME move makes the estimate"
        " worse"
    )
    print("   pi_sense   mu_post   sd_post   |mu_post| vs signal")
    for pi_s in (4.0, 2.0, 1.0, 0.5, 0.25):
        m, p = combine(3.0, 1.0, Y, pi_s)
        print("   {:8.2f} {:9.3f} {:9.3f}   {:5.0f}% of the signal"
              .format(pi_s, m, 1 / math.sqrt(p), 100 * m / Y))

    print("\nC. what has to change for the posterior to come down:")
    base_m, base_p = combine(3.0, 1.0, Y, 2.0)
    print(
        "   baseline (alarmed prior, pi_sense 2): mu_post = {:.3f}".format(
            base_m
        )
    )
    for lab, args in (
        ("halve sensory precision", (3.0, 1.0, Y, 1.0)),
        ("halve PRIOR precision", (3.0, 0.5, Y, 2.0)),
        ("move the prior to 0", (0.0, 1.0, Y, 2.0)),
        ("move prior to 0 AND halve its precision", (0.0, 0.5, Y, 2.0)),
    ):
        m, p = combine(*args)
        print(
            "   {:42s} mu_post = {:+.3f}  ({:+.3f})".format(
                lab, m, m - base_m
            )
        )

    print(
        "\nD. cost of the shift, KL from the alarmed to the revised posterior"
        " (nats):"
    )
    for lab, args in (("halve sensory precision", (3.0, 1.0, Y, 1.0)),
                      ("move the prior to 0", (0.0, 1.0, Y, 2.0))):
        m, p = combine(*args)
        print("   {:42s} {:.4f}".format(lab, kl_gauss(m, p, base_m, base_p)))
# A. benign prior (mu_prior = 0): lowering sensory precision helps
#    pi_sense   mu_post   sd_post   |mu_post| vs signal
#        4.00     1.600     0.447      80% of the signal
#        2.00     1.333     0.577      67% of the signal
#        1.00     1.000     0.707      50% of the signal
#        0.50     0.667     0.816      33% of the signal
#        0.25     0.400     0.894      20% of the signal
#
# B. alarmed prior (mu_prior = 3): the SAME move makes the estimate worse
#    pi_sense   mu_post   sd_post   |mu_post| vs signal
#        4.00     2.200     0.447     110% of the signal
#        2.00     2.333     0.577     117% of the signal
#        1.00     2.500     0.707     125% of the signal
#        0.50     2.667     0.816     133% of the signal
#        0.25     2.800     0.894     140% of the signal
#
# C. what has to change for the posterior to come down:
#    baseline (alarmed prior, pi_sense 2): mu_post = 2.333
#    halve sensory precision                    mu_post = +2.500  (+0.167)
#    halve PRIOR precision                      mu_post = +2.200  (-0.133)
#    move the prior to 0                        mu_post = +1.333  (-1.000)
#    move prior to 0 AND halve its precision    mu_post = +1.600  (-0.733)
#
# D. cost of the shift, KL from the alarmed to the revised posterior (nats):
#    halve sensory precision                    0.0889
#    move the prior to 0                        1.5000

Section A is the case in which the original claim holds: with a prior of 0, lowering sensory precision from 4.0 to 0.25 brings the posterior from 80 % of the signal down to 20 %. Section B is the same operation under an alarmed prior (μ0=3.0\mu_0 = 3.0): the posterior rises from 110 % to 140 % of the signal, that is, from 2.200 to 2.800. Section C sets four interventions side by side: lowering sensory precision raises the posterior by 0.167; moving the prior to 0 lowers it by 1.000. Section D gives the Kullback–Leibler distance between the posteriors for the two moves: 0.0889 against 1.5000, a factor of almost seventeen.

The posterior mean against the precision of the interoceptive channel, \pi_y. Left: with a benign prior (\mu_0 = 0, dotted line), lowering precision moves the posterior (solid) down from the signal value (gold dashed). Right: with an alarmed prior (\mu_0 = 3), the same move carries the posterior upward. The arrows mark the direction in which the posterior moves as precision falls.

This is the dark-room problem of the free-energy framework: if the goal were only to reduce prediction error, the optimal policy would be to hide in a dark room where nothing happens (Friston, Thornton, and Clark 2012). Friston, Thornton and Clark reply that the prior itself contains expectations of varied sensory input, so the dark room is not really a free-energy minimum. But that reply makes precisely this chapter’s point: what decides the outcome is the prior, not the size of the sensory precision.

Hypothesis · not established

The author’s hypothesis (not established): the active ingredient in attending to the body without interpreting it is not a reduction in the precision of the interoceptive channel but a reduction in the alarm of the prior itself; while the prior remains alarmed, the same practice raises the subjective estimate of bodily threat, and presents as symptoms worsening early in practice. What would show it to be wrong: in a pre-registered trial stratified by a baseline measure of the “bodily threat prior” (body vigilance or anxiety sensitivity, for instance), finding improvements of the same direction and similar size in the high-alarm and low-alarm strata would refute it.

A two-level model: what is decoupled

Those two expressions are static. The original claim has a dynamic component as well: “higher-order awareness soothes the default mode network”, that is, the estimate about the body stops driving the estimate about the situation. Written as a two-level predictive-coding loop: the first level holds a bodily state estimate x1x_1, the second a situation estimate x2x_2 (“is this dangerous?”), and a scalar πh\pi_h scales the whole second-level pathway.

dx1dt=k(π1(y−x1)−πhπ2(x1−x2)),dx2dt=kπhπ2(x1−x2)−λx2.\frac{dx_1}{dt} = k\big(\pi_1 (y – x_1) – \pi_h \pi_2 (x_1 – x_2)\big), \qquad \frac{dx_2}{dt} = k\,\pi_h \pi_2 (x_1 – x_2) – \lambda x_2 .

Calculation

# Chapter 1. A two-level predictive-coding loop with an explicit higher-order
# precision term, integrated with the Euler method (dt = 0.001, 20 s). Level 1
# holds a bodily state estimate x1 driven by an interoceptive sample y; level
# 2 holds a context estimate x2 ("is this situation dangerous?").
# Precision-weighted errors:
#   e1 = y - x1,            e2 = x1 - x2
#   dx1/dt = k (pi_1 e1 - pi_2 e2),   dx2/dt = k pi_2 e2 - lambda x2
# pi_h scales the WHOLE second-level pathway (pi_2 -> pi_h pi_2): this is the
# one knob the author's framework calls "higher-order awareness". The run
# shows what it does to the steady state and to the time constant. Author's
# model; illustrative dimensionless units.
import math

DT, T = 0.001, 20.0

def run(pi1, pi2, pi_h, y=2.0, k=4.0, lam=0.25):
    x1 = x2 = 0.0
    peak_rate = 0.0
    t_half, target = None, None
    traj = []
    n = int(T / DT)
    for i in range(n):
        e1, e2 = y - x1, x1 - x2
        d1 = k * (pi1 * e1 - pi_h * pi2 * e2)
        d2 = k * pi_h * pi2 * e2 - lam * x2
        x1 += DT * d1
        x2 += DT * d2
        peak_rate = max(peak_rate, abs(d2))
        traj.append((i * DT, x1, x2))
    # steady state from the last 1 s
    tail = traj[-int(1.0 / DT):]
    ss1 = math.fsum(r[1] for r in tail) / len(tail)
    ss2 = math.fsum(r[2] for r in tail) / len(tail)
    # time for x2 to reach 63 % of its steady state
    if abs(ss2) > 1e-9:
        target = 0.632 * ss2
        for t, _, v in traj:
            if (v >= target) if ss2 > 0 else (v <= target):
                t_half = t
                break
    return ss1, ss2, t_half, peak_rate, traj

if __name__ == "__main__":
    print("pi_h   x1*      x2*     tau63(s)   peak |dx2/dt|")
    for pi_h in (2.0, 1.0, 0.5, 0.25, 0.1, 0.0):
        s1, s2, th, pk, _ = run(1.0, 1.0, pi_h)
        ts = "{:8.3f}".format(th) if th is not None else "       -"
        print(
            "{:4.2f} {:7.3f} {:8.3f} {} {:14.3f}".format(pi_h, s1, s2, ts, pk)
        )

    print(
        "\nThe second level is what carries the interoceptive signal into a"
        " context"
    )
    print(
        "estimate. Lowering pi_h lowers x2* and slows the approach; at pi_h ="
        " 0 the"
    )
    print(
        "second level is cut off and x2* = 0, while x1* tracks the body"
        " unchanged."
    )

    print(
        "\nSensitivity of x2* to the interoceptive sample y (pi_h = 1 vs"
        " 0.25):"
    )
    print("   y      x2*(pi_h=1)   x2*(pi_h=0.25)   ratio")
    for y in (0.5, 1.0, 2.0, 4.0):
        _, a, _, _, _ = run(1.0, 1.0, 1.0, y=y)
        _, b, _, _, _ = run(1.0, 1.0, 0.25, y=y)
        print(
            "   {:4.1f} {:13.3f} {:16.3f} {:7.2f}".format(
                y, a, b, b / a if a else float("nan")
            )
        )

    print(
        "\nNOTE: x1* is unchanged by pi_h in this model, so 'decoupling' here"
    )
    print(
        "means the context estimate stops tracking the body, NOT that the"
        " body"
    )
    print("signal goes away.")
# pi_h   x1*      x2*     tau63(s)   peak |dx2/dt|
# 2.00   1.886    1.829    0.580          2.689
# 1.00   1.889    1.778    0.688          2.139
# 0.50   1.895    1.684    0.883          1.558
# 0.25   1.905    1.524    1.209          1.033
# 0.10   1.926    1.185    1.886          0.526
# 0.00   2.000    0.000        -          0.000
#
# The second level is what carries the interoceptive signal into a context
# estimate. Lowering pi_h lowers x2* and slows the approach; at pi_h = 0 the
# second level is cut off and x2* = 0, while x1* tracks the body unchanged.
#
# Sensitivity of x2* to the interoceptive sample y (pi_h = 1 vs 0.25):
#    y      x2*(pi_h=1)   x2*(pi_h=0.25)   ratio
#     0.5         0.444            0.381    0.86
#     1.0         0.889            0.762    0.86
#     2.0         1.778            1.524    0.86
#     4.0         3.556            3.048    0.86
#
# NOTE: x1* is unchanged by pi_h in this model, so 'decoupling' here
# means the context estimate stops tracking the body, NOT that the body
# signal goes away.

Three things are worth recording. First, as πh\pi_h falls, the steady state of the situation estimate x2*x_2^* drops from 1.778 to 1.185 (at πh=0.1\pi_h = 0.1) and to 0 at πh=0\pi_h = 0; while the bodily estimate x1*x_1^* barely changes (1.889 to 2.000). So in this model “decoupling” means that the situation estimate stops tracking the body, not that the bodily signal disappears. The practice does not make the heartbeat imperceptible; it stops the heartbeat being read as danger. This is the same thing as the “non-judgmental” of Chapter 3, written another way.

Second, as πh\pi_h falls the time constant of the approach to steady state lengthens (0.688 s to 1.886 s). The situation estimate in this model is not only lower but slower. Slowness is a benefit for harmless bodily events and not for real dangers; Chapter 3 makes that precise with a signal-detection model.

Third, the sensitivity of x2*x_2^* to the bodily signal yy stands in a fixed ratio between the two values of πh\pi_h (0.86 in all four rows). That is because the model is linear in yy. Real systems are not, but it shows something: in a linear model, “lowering higher-order precision” is a gain operation and not a threshold operation; it shrinks every situational response in proportion, without distinguishing the responses that should shrink from those that should not.

Note

πh\pi_h is not a known physiological quantity. This book treats πh\pi_h as a model parameter and does not claim it corresponds to any measurable neural signal. Identifying precision-weighting with a particular neural mechanism — the power in some frequency band, or the gain of some pathway — requires independent evidence, and Volume 8, Chapter 3 noted that such identifications are often made prematurely in the entropic-brain literature.

Measuring interoception: a problem that has to be solved first

To test any claim about interoception, interoception has to be measurable. There is a methodological trap here.

Evidence

The problem with heartbeat counting. “Interoceptive accuracy” has long been measured with the heartbeat counting task: participants count their heartbeats over an interval without taking a pulse, and the count is compared with the true number. Zamariola and colleagues identify fundamental problems with the resulting score: it correlates strongly with actual heart rate, participants who undercount can score higher, and simple bivariate correlations reveal these confounds (Zamariola et al. 2018). Garfinkel and colleagues had earlier distinguished three levels — interoceptive accuracy (objective performance), interoceptive sensibility (self-reported disposition) and interoceptive awareness (metacognition, the correspondence between performance and confidence) — which correlate only weakly (Garfinkel et al. 2015). The Multidimensional Assessment of Interoceptive Awareness measures self-reported styles of attending, including subscales such as “not-worrying” and “not-distracting” (Mehling et al. 2012), a different construct from objective performance.

This has a direct consequence for Chapter 8: the trial in this book uses heartbeat counting as no outcome at all. What the original claim is about is a person’s relation to bodily signals, and that is what the self-report scales measure, not what heartbeat counting measures. Conflating the two produces uninterpretable results: a practice might raise scores on “I attend to my body more easily” while leaving heartbeat counting performance completely unchanged, and neither of those is “interoception became more accurate”.

Evidence

Interoception and mental health (review). Paulus and Stein proposed that interoceptive abnormality in anxiety and depression can be written as an amplification of bodily prediction signals (Paulus and Stein 2010). Khalsa and colleagues’ roadmap paper reviews the field and states its disagreements about measurement and definition explicitly (Khalsa et al. 2018). This is an active field whose methods are not yet settled; this book cites it as the source of the questions, not as the basis of the conclusions.

Summary

Interoception has well-described anatomical pathways, and writing it as precision-weighted inference is a productive framework. But the framework’s first expression refutes the first link of the original claim: lowering the precision of the interoceptive channel pushes the posterior toward the prior, so it lowers the estimate of bodily threat only when the prior is benign, and raises it when the prior is alarmed (program p01: from 2.333 to 2.800). Bringing the estimate down requires moving the prior, which is a different operation from declining to interpret, and about seventeen times more costly in distributional terms. The two-level model shows that what “decoupling” amounts to is the situation estimate ceasing to track the bodily estimate, not the bodily signal disappearing, and that it simultaneously makes the situation estimate slower (time constant from 0.688 s to 1.886 s). Finally, the measurement of interoception has unsolved problems of its own: heartbeat counting scores cannot be taken as interoceptive accuracy, and self-report scales measure a different construct from objective performance; Chapter 8 chooses its outcomes accordingly.

References

Barrett, Lisa Feldman, and W. Kyle Simmons. 2015. “Interoceptive Predictions in the Brain.” Nature Reviews Neuroscience 16: 419–29. https://doi.org/10.1038/nrn3950.
Craig, A. D. 2002. “How Do You Feel? Interoception: The Sense of the Physiological Condition of the Body.” Nature Reviews Neuroscience 3: 655–66. https://doi.org/10.1038/nrn894.
———. 2009. “How Do You Feel — Now? The Anterior Insula and Human Awareness.” Nature Reviews Neuroscience 10: 59–70. https://doi.org/10.1038/nrn2555.
Critchley, Hugo D., and Neil A. Harrison. 2013. “Visceral Influences on Brain and Behavior.” Neuron 77 (4): 624–38. https://doi.org/10.1016/j.neuron.2013.02.008.
Critchley, Hugo D., Stefan Wiens, Pia Rotshtein, Arne Öhman, and Raymond J. Dolan. 2004. “Neural Systems Supporting Interoceptive Awareness.” Nature Neuroscience 7: 189–95. https://doi.org/10.1038/nn1176.
Feldman, Harriet, and Karl J. Friston. 2010. “Attention, Uncertainty, and Free-Energy.” Frontiers in Human Neuroscience 4: 215. https://doi.org/10.3389/fnhum.2010.00215.
Friston, Karl. 2010. “The Free-Energy Principle: A Unified Brain Theory?” Nature Reviews Neuroscience 11: 127–38. https://doi.org/10.1038/nrn2787.
Friston, Karl, Christopher Thornton, and Andy Clark. 2012. “Free-Energy Minimization and the Dark-Room Problem.” Frontiers in Psychology 3: 130. https://doi.org/10.3389/fpsyg.2012.00130.
Garfinkel, Sarah N., Anil K. Seth, Adam B. Barrett, Keisuke Suzuki, and Hugo D. Critchley. 2015. “Knowing Your Own Heart: Distinguishing Interoceptive Accuracy from Interoceptive Awareness.” Biological Psychology 104: 65–74. https://doi.org/10.1016/j.biopsycho.2014.11.004.
Khalsa, Sahib S., Ralph Adolphs, Oliver G. Cameron, Hugo D. Critchley, Paul W. Davenport, Justin S. Feinstein, Jamie D. Feusner, et al. 2018. “Interoception and Mental Health: A Roadmap.” Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 3 (6): 501–13. https://doi.org/10.1016/j.bpsc.2017.12.004.
Mehling, Wolf E., Cynthia Price, Jennifer J. Daubenmier, Mike Acree, Elizabeth Bartmess, and Anita Stewart. 2012. “The Multidimensional Assessment of Interoceptive Awareness (MAIA).” PLOS ONE 7 (11): e48230. https://doi.org/10.1371/journal.pone.0048230.
Paulus, Martin P., and Murray B. Stein. 2010. “Interoception in Anxiety and Depression.” Brain Structure and Function 214: 451–63. https://doi.org/10.1007/s00429-010-0258-9.
Petzschner, Frederike H., Lilian A. E. Weber, Tim Gard, and Klaas E. Stephan. 2017. “Computational Psychosomatics and Computational Psychiatry: Toward a Joint Framework for Differential Diagnosis.” Biological Psychiatry 82 (6): 421–30. https://doi.org/10.1016/j.biopsych.2017.05.012.
Rao, Rajesh P. N., and Dana H. Ballard. 1999. “Predictive Coding in the Visual Cortex: A Functional Interpretation of Some Extra-Classical Receptive-Field Effects.” Nature Neuroscience 2: 79–87. https://doi.org/10.1038/4580.
Seth, Anil K., and Karl J. Friston. 2016. “Active Interoceptive Inference and the Emotional Brain.” Philosophical Transactions of the Royal Society B 371 (1708): 20160007. https://doi.org/10.1098/rstb.2016.0007.
Uddin, Lucina Q. 2015. “Salience Processing and Insular Cortical Function and Dysfunction.” Nature Reviews Neuroscience 16: 55–61. https://doi.org/10.1038/nrn3857.
Zamariola, Giorgia, Pierre Maurage, Olivier Luminet, and Olivier Corneille. 2018. “Interoceptive Accuracy Scores from the Heartbeat Counting Task Are Problematic: Evidence from Simple Bivariate Correlations.” Biological Psychology 137: 12–17. https://doi.org/10.1016/j.biopsycho.2018.06.006.

Fan, Y. (2026). Interoceptive Decoupling, Salience-Network Switching and Autonomic Visceral Restoration: From Precision-Weighting in the Default Mode Network and the Redistribution of Vagal Tone to the Neurophenomenology of a Non-Judgmental Imagery Stream, with a Textual Examination of neiguan in the Huangdi Neijing [内感受解离、突显网络切换与内脏自主神经修复] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24020418 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).