Pain suffering and the self.

An active allostatic inference

explanation
Philip Gerrans*

Department of Philosophy, University of Adelaide, Adelaide, SA, Australia

If the self is an illusion it is a pretty painful one.
*Corresponding author. Department of Philosophy, University of Adelaide, Adelaide, Australia. E-mail: [email protected]

Abstract
Distributed processing that gives rise to pain experience is anchored by a multidimensional self-model. I show how the phenomenon
of pain asymbolia and other atypical pain-related conditions (Insensitivity to Pain, Chronic Pain, ‘Social’ Pain, Insensitivity to Pain,
Chronic Pain, ‘Social’ Pain, empathy for pain and suffering) can be explained by this idea. It also explains the patterns of association
and dissociation among neural correlates without importing strong modular assumptions. It treats pain processing as a species of
allostatic active inference in which the mind co-ordinates its processing resources to optimize basic bodily functioning at different
time scales. The self is inferred to be source and target of regulation in this process. The self-modelling account reconciles conflicting
deaffectualization and depersonalization accounts of pain asymbolia by showing how depersonalization and pain asymbolia arise at
different levels of hierarchical self modelling.

Keywords: active inference; pain asymbolia; self model; Allostasis; pain disorders; affect

Introduction In the remainder of this paper I use the related concepts of

allostatic active inference and self modelling to develop a neuro-
Intuitively pain is an experience of bodily damage that is felt to
computational account of pain asymbolia. Pain processing turns
be unpleasant and aversively motivating. A property of pain that
out to be a particular instance of the general phenomenon of allo-
has interested philosophers is its intimate connection with self-
awareness. As Wittgenstein said ‘I can only believe that someone static active inference anchored by a self model. An advantage
else is in pain, but I know [immediately and non-inferentially] if of this account is its ability to explain the intuitive conception of
I am’. On this view pain is a condition which is essentially felt to pain as well as various disorders in which typical features of pain
me ‘mine’. episodes dissociate.
An integrative theory of pain explains how neurocomputa- I first describe pain asymbolia and a variety of conditions
tional processes give rise to the characteristic phenomenology. which pain sensation, distress, aversive motivation, and ‘mine-
Such theories draw on neuroscience, neuropsychology, clinical ness’ dissociate. I describe how the neuroscience of pain process-
practice, and the psychology of pain management. Philosophers ing accommodates this complexity by treating pain processing in
have recently incorporated this type of evidence to refine philo- terms of either a modular or a matrix architecture. In the mod-
sophical theories of pain. Of particular interest is a disorder ular conception, pain processing has discrete, dissociable, com-
called pain asymbolia that raises questions about the relationship ponents responsible for nociception (detection transduction and
between nociception (transduction of sensory signals of damage transmission of signals of threat to bodily integrity) and affective
or threat to bodily integrity), pain sensation, affect, and motiva- processing. This leads to an elegant ‘deaffectualization’ explana-
tion. Patients with pain asymbolia consequent on damage to the tion of pain asymbolia in which affective processing is impaired
posterior insula cortex say that they are in (often intense) pain but but nociceptive processing is intact. So the patient still feels the
they are not distressed by that pain and they are not motivated to nociceptive aspect of pain but is not distressed by it and con-
avoid the noxious stimulus. It is therefore an empirical challenge sequently not motivated to avoid it. However, dissociations at
to the intuitive conception of pain. The challenge is increased by either the level of neurocircuitry or phenomenology are not as
reports of pain asymbolia in which subjects report that it feels as clean as predicted by a ‘strictly’ modular componential process-
if the pain does not belong to them. On some interpretations sub- ing architecture for pain. And, once we examine a broader range of
jects with pain asymbolia have lost the sense of ‘mineness’ for pain-related phenomena such as chronic pain, congenital insen-
pain. ‘Mineness’ is a philosophical term of art that refers to the sitivity to pain, distress, and empathy in different conditions, the
feeling that experiences belong to the subject (Billon 2016, 2023). modular theory of pain processing that underlies the affective

© The Author(s) 2024. Published by Oxford University Press.

This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
2 Gerrans

theory of pain asymbolia breaks down. So insofar as the deaffec- cortices. The posterior insula is a hub of processing that integrates
tualization theory relies on a modular processing architecture it basic bodily signals including nociception to produce the feeling of
requires revision. embodied selfhood (‘material me’ as Seth (2013) calls it). Activity
Problems with a philosophical version of the deaffectualization in the AI links activity in the posterior insula to activity in circuitry
theory (Grahek 2011) led Colin Klein (2015) to propose a revised that implements emotional and cognitive processes. The resultant
account. He proposed that that instead of thinking of asymbo- interaction between bodily, emotional, and cognitive processing
lia as a phenomenon of deaffectualization we should think of it allows ‘raw’ bodily signals to be transcribed into affective signals
as a species of ‘depersonalization for pain’. This fits with reports that inform the subject not just that her body is in a particular
of depersonalization that resemble reports of asymbolia, like the state but the significance of that state for ‘her’. For example, an
following. interoceptive signal of bodily arousal can be experienced as exhil-
aration, anxiety, arousal, or part of an episode of anger depending
When a part of my body hurts, I feel so detached from the pain on the context. I now apply these ideas to explain pain asymbolia.
that it feels as if it were somebody else’s pain (Sierra 2009).

Pain asymbolia. Sensation and affect

I endorse the conceptual revision advocated by Klein’s deperson-
alization interpretation but it faces related problems. The most Pain asymbolia is a rare condition in which patients report a sense
persuasive accounts of depersonalization are (i) modular deaffec- of detachment from their experience of pain.
tualization accounts that (ii) explain deaffectualization in terms
of hypoactivity in the ‘anterior’ insula cortex, hypothesized to be ‘In the absence of primary sensory deficits’, these six patients
a neural correlate of distress. But the neural correlate of pain showed a lack of withdrawal behaviours and absence of emo-
asymbolia is impaired functioning in the ‘posterior’ insula cortex. tional reactions (or inappropriate emotional responses) to
I propose a solution that (i) explains the patterns of associ- painful stimuli applied to the whole body, as well as to threaten-
ation and dissociation between pain, motivation, and affect in ing gestures. Five patients also did not react to verbal threats.
pain asymbolia and other pain disorders; (ii) explains the rela- The ‘patients did not appear concerned about the defect and
tionship between basic bodily (such as nociceptive) and affective seemed incapable of learning appropriate escape or protective
experience; and (iii) explains the roles of neural substrates of behaviors’ (Berthier et al. 1988).
pain asymbolia (posterior insula) and depersonalization (anterior
insula [AI]) and their relationship.
There are two components to the explanation. The first is the The experience is not of insensitivity to bodily damage but rather
idea that pain phenomenology is an emergent result of process- the feeling that the sensation of bodily damage does not matter
ing across distributed circuitry baptized the ‘pain matrix’. This to the patient. Sometimes it is described as feeling pain without
matrix is configured in different ways under different conditions, caring about it or being motivated to avoid it. It has also been com-
giving rise to the different disorders of pain processing. The idea pared to feelings of dissociation or anaesthesia. This is consistent
that pain processing depends on activity in a distributed neuro- with the classic case reported by Schilder and Stengel.
matrix is fairly standard but its implications for understanding
disorders of pain processing are not fully worked out. This paper The patient displays a striking behaviour in the presence of
is a contribution to that project. pain. She reacts either not at all or insufficiently to being
The second is the idea that pain processing is a form of allo- pricked, struck with hard objects, and pinched. She never pulls
static active inference. The concept of allostasis extends the con- her arm back energetically or with strength. She never turns the
cept of homeostasis to encompass diverse forms of regulatory torso away or withdraws with the body as a whole. She never
activity integrated to optimize organismic functioning (Barrett attempts to avoid the investigator (Schilder and Stengel 1931).
et al. 2016, Corcoran and Hohwy 2017, Kleckner et al. 2017). The
concept of active inference is a corollary of allostasis. Active infer-
Classic neuropsychological accounts (sensorimotor-limbic discon-
ence treats the mind as [solving] ‘biological regulation problems
nection accounts as they are known) treated pain asymbolia as
by learning internal (generative) models of their bodily and inte-
evidence for the idea that pain processing has a modular archi-
roceptive processes’ (Tschantz et al. 2022, 108266). In the case of
tecture with a ‘sensory’ and an ‘affective’ component (Mesulam
pain this means that the brain infers that the self is the thing
2000). This wave of theorizing about pain asymbolia is anchored
that senses damage, initiates appropriate internal (regulatory) and
in neuroscientific explanation of pain, particularly conditions like
external (avoidance) responses, and feels the consequences. To
chronic pain, placebo analgesia, congenital insensitivity for pain,
act adaptively in the case of pain many disparate processes rang-
and empathy for pain. Each of these phenomena has been ele-
ing from opoid release and cytokine cascade (low-level dynamical
gantly explained on the assumption that ‘sensory’ pain and ‘affec-
processes) to the evaluation of nociceptive signals (an emotional
tive’ or ‘emotional’ pain are produced by distinct components of a
process that gives rise to the distressing aspect of pain) to consult-
specialized pain processing system. On one version of this modu-
ing a Harley St specialist or tiktok influencer must be co-ordinated.
lar view, pain asymbolia is the result of loss of affective response
The brain solves this problem by attributing signals arising in dif-
to noxious stimuli. This would explain why asymbolic subjects do
ferent modalities to the same unified persisting entity: the self
not react aversively to pain or experience it as distressing, even
(Moutoussis et al. 2014). This allows a complex system to be reg-
though they continue to sense damage to their body.
ulated as a single entity. When the circuitry that implements this
‘self model’ is identified and its role described, some of the puz-
zling aspects of pain phenomenology are clarified. This self model The syndrome of asymbolia to pain appears to be a
is implemented by a pattern of neural activity that integrates bod- somatosensory-limbic disconnection syndrome. Pain asymbo-
ily emotional/affective and narrative/conceptual processing. Key lia can be conceptualized as a somatosensory analogue of
circuits that sustain this pattern are the posterior insula and AI visual ‘hypoemotionality’ (Sierra et al. 2002).
 Pain suffering and the self. 3

This modular explanation of pain processing and pain asymbolia activations in the amygdala, AI, and cingulate cortices, ‘which
accommodates a consensus about its neural substrates. Nocicep- are regions implicated in the affective aspects of pain’, are max-
tive input is processed via distinct (but not discrete) neuroanatom- imally suppressed at the lowest opioid dose (Lee et al. 2014,
ical pathways. A sensorimotor pathway conveys nociceptive infor- Rütgen et al. 2015).
mation to the somatosensory cortex and insula. Another projects
to the insula, the paralimbic and limbic systems, and ultimately This is consistent with the idea that the AI-ACC circuit is impli-
prefrontal cortex. At the same time, activity in these pathways cated in personal distress. This form of suffering can be produced
can be influenced by descending modulatory systems that target and regulated relatively independently of low-level nociception
different aspects of the system at different time scales. A circuit and reflexive behavioural responses like flinching and withdrawal.
linking the AI and rostral anterior cingulate cortex plays a crucial These studies suggest that pain processing occurs along dif-
role in sustaining the feeling of distress and linking it to higher ferent dimensions with sensorimotor responses occurring earlier
level (narrative and conceptual) forms of self-awareness. Thus, and outside the scope of deliberate control. A primary hub of this
there is preliminary evidence that the distress (‘souffrance’) of processing is the ‘posterior insula’, consistent with its role in inte-
pain and the sensation of bodily damage (‘douleur’) are potentially grating low-level bodily signals crucial in maintaining organismic
dissociable because their substrates, although they communicate viability. Threats to bodily integrity detected in nociception are
in the construction of the pain experience, are distinct (Berthier processed by the posterior insula since they are vital to homeo-
et al. 1988, Danziger 2006, Starr et al. 2009, Klein 2015, Gerrans static regulation (Gu et al. 2013, Frot et al. 2014). It is crucial to
2020). emphasize this point because the posterior insula is ‘not’ hypoth-
esized to be the substrate of the experience of distress but of an
earlier stage of pain processing that integrates the nociceptive sig-
Evidence for dissociations nal with other bodily signals as part of a system of basic bodily
In support of the idea of distinct and dissociable pathways for regulation.
sensorimotor and affective processing, the modular theorist can The feelings of suffering and distress that are part of the pain
point to conditions such as chronic pain. Chronic pain is explained experience are produced at a higher level or later stage of process-
as a form of neuroplastic change in which circuitry involved ing that involves the AI, modulated by prefrontal cortical struc-
in affective processing ‘takes over’. Such cases are described as tures that represent personal and social information and explicit
‘nociplastic’ rather than nociceptive. narrative and conceptual self-knowledge. If this is correct, then
one can see the appeal of the modular account. On the assump-
According to a recent meta-analysis, the experimental induc- tion that the mechanisms that support these distinct aspects of
tion of acute pain (e.g. with painful versus non-painful thermal processing are relatively specialized and can be selectively acti-
stimuli) is generally associated with activations of both sen- vated or damaged, one can see why it makes sense to partition
sory (e.g. thalamus, secondary somatosensory cortices (SII), pain processing into sensori-motor and affective-personal com-
dorsal posterior insula) and affective (e.g. dorsal anterior cin- ponents. However, such an account needs to explain why pain
gulate cortex [ACC], AI) brain regions in healthy adults and pain asymbolia is associated with damage to the posterior rather than
patients (Xu 2020). AI. After all, on the modular account, it is the activity in an AI that
produces feelings of distress. So one might predict that pain asym-
bolia would be the result of damage/dysfunction in anterior rather
In contrast, nociplastic pain conditions, such as nonspecific
than posterior insular regions. I return to this question below.
chronic low back pain (cLBP), are associated with altered neural
activation patterns in ‘affective brain regions only’, particularly
the rostral ACC, mPFC, and amygdala (Gu et al. 2013). Empathy and chronic pain
The modular deaffectualization account can point to further evi-
This is consistent with the classic modular conception in which dence that the AI is the substrate of distressing experience. Some
the primary substrate for suffering is a circuit centred on the of the most dramatic evidence comes from studies of patients who
AI and anterior cingulate cortices. Indeed, a last ditch treat- are congenitally insensitive to pain and patients with congenital
ment for chronic pain is lobotomy that disconnects this circuit agenesis of body parts. Such patients clearly lack early sensory
from prefrontal regions. Patients treated by lobotomy reported processing of bodily damage, either because they lack activity in
that the intensity and nature of the pain had not changed and the necessary sensorimotor processing circuitry or lack the body
their automatic responses to a noxious stimulus were ‘intact and part itself. Nonetheless, when viewing images of bodily damage,
often exaggerated’. However, they are untroubled by pain or the the patients experience distress. This can be described as a form of
prospect thereof. As Danziger (2006) puts it ‘[E]motional impact of empathy where empathy is taken to mean the ability to experience
chronic pain is dramatically reduced’. affective states while observing, but not actually experiencing,
Cases such as this form part of sensory elicitation of those states. i.e. seeing or imaging someone
else (or oneself) in pain or distress.

a wealth of evidence suggesting that patients with chronic

Viewing body parts in pain corresponding to the missing limb
pain may have anatomical alterations within regions involved
induces a significant activation ‘only in brain regions devoted
in cognitive and emotional modulation of pain, such as the
to emotional empathy, such as the anterior insula’ (Betti and
dorsolateral and medial PFC, the ACC, and the insula (Barrett
Aglioti 2016).
2017).
Interestingly, however, in the absence of a somatic representation,
Another consideration that supports this is the action of i.e. perception of the body part, ‘the understanding of another’s
morphine analgesia. In low doses, the primary target of pain relies on inferential mechanisms rather than affective reso-
morphine is nance mapping’ (Betti and Aglioti 2016, 195). The idea behind this
4 Gerrans

enigmatic sentence is that empathetic distress in the absence of substrates of these systems are circuits involving posterior insula
nociception can be driven by higher level cognition, for example and AI, respectively. On this view, Pain Asymbolia and Chronic
by thinking about others’ painful experiences. Neuroplastic Pain represent a form of double dissociation. Asym-
The authors of this review conclude that both perception of bolia is evidence of preserved sensory processing and absent
other’s pain and experience of pain are typically initially processed affective processing, and Chronic Neuroplastic Pain is evidence
by sensorimotor structures and subsequently processed by an of reduced sensory processing and intact or exaggerated affective
affective system that evaluates the significance of bodily damage processing.
for the subject. However, in the absence of initiating sensorimotor Affective states are typically thought of as ‘intrinsically’ moti-
processing, the affective system remains able to produce a feeling vational. We avoid distressing situations and approach pleasant
of distress for a perceived or inferred injury ‘to self or other’. ones. This would explain why patients with pain asymbolia, who
Tania Singer and collaborators asked female members of cou- are not distressed by pain sensations, lack aversive motivation.
ples to observe their partners experiencing a mild electrical shock However, as Klein (2015) among others points out, if affect is
to the wrist. They were not explicitly instructed to empathize, only intrinsically motivating, lobotomized patients who have flattened
to observe. Patterns of neural activation were compared to a condi- affective responses should lose their aversive reaction to noxious
tion in which they received a shock themselves. The key difference stimuli. However, their reflexive aversive responses are intact or
reported was that exaggerated (Danziger 2006; Duquette et al. 2007). So, ‘there is no
necessary connection between affective experience and aversive
only part of the network mediating pain experience is shared behaviour’.
when empathizing with pain in others. Empathizing with some- Similarly, patients with chronic insensitivity to pain have intact
one else’s pain elicited activity principally in left and right AI, affective circuitry but they have no motivation to avoid painful
ACC, lateral cerebellum, and brainstem (Singer et al. 2004). stimuli, typically with disastrous results. ‘She reported numerous
burns and cuts without pain (Supplementary Fig. S1), often smelling
The main contrast with the ‘self’ pain condition was that her burning flesh before noticing any injury’, Br J Anaesth. 2019 Aug;
123(2): e249–e253.
pain-related activation in contralateral SI, SII/posterior insula,
and caudal ACC are specific to self-experienced pain, as
opposed to perceived pain in others.
Against the modular theory of pain
asymbolia
Their conclusion is a very clear statement of the view that Cases like these suggest that processing of the nociceptive signal
pain processing has dissociable sensorimotor/discriminative and represents the brain’s response to the significance of the stim-
affective components and it is the latter that are involved in ulus in that context, with sensory, affective, behavioural, and
empathy. cognitive processes all contributing. However, in non-standard
contexts, elements of this processing system can dissociate in
(Zhou et al. 2020) Rostral ACC and AI appear to reflect the emo- different ways. But this does not license the modular hypoth-
tional experience that evokes our re-actions to pain and consti- esis of discrete dissociable components of a specialized pain
tutes the neural basis for our understanding of the feelings of processing system. When we turn to a wider range of evi-
others and ourselves. (Singer et al. 2004 1161, my italics). dence about pain processing, we see ‘interdependence rather
than independence’ of sensorimotor and affective aspects of pain
that connectivity between mPFC and regions of the salience processing.
network (SN), including the insula and ACC, were found to be One illustration of this coupling is provided by empathetic
increased among patients with nociplastic pain compared to response in different conditions. Recall that Singer’s experi-
controls (Yarns et al. 2022 104558). ments used couples observing a partner receiving a low-intensity
painful stimulus. In that condition, the circuitry activated by
(Wang et al. 2020) The idea that an ACC-AI circuit is the sub- both observation and personal experience was the AI/rostral ACC,
strate for feelings of distress is also part of the explanation of the hypothesized basis of distress. However, in other conditions
chronic pain (Simons 2014). For example, a meta-analysis of stud- in which subjects view severe injuries or wounded body parts,
ies comparing patients with and without chronic pain reported a somatosensory aspects of the pain processing system are also
consistent role of the AI consistent with its ‘complex role in pro- activated (Betti and Aglioti 2016). This is consistent with some
cesses directly or indirectly related to the acute and chronic pain somatosensory resonance or contagion conceptions of empathy
experience, including pain empathy (Fallon et al. 2020; Xu 2020; that emphasize that third person observation and first person
Zhou et al. 2020), interoception and salience processing (Li et al. experience of body state can activate the same sensory processing
2020; Yao et al. 2018) as well as emotional experience (Gogolla systems (Singer and Lamm 2009).
2017; Ferraro et al. 2022). Another study of emotional regulation The case of ‘social pain’ evoked by ostracism or criticism makes
and pain processing found (Duquette et al. 2007; Yao et al. 2018). a similar point. It is natural to conceive of it in terms of affec-
As a consequence, the most effective interventions for chronic tive processing: the distress evoked by rejection. As such, on
pain do not target nociceptive processing pathways or posterior a strictly modular view, one might predict activity in the AI-
insula where they converge and are integrated with other basic ACC associated with affective experience, and this is confirmed.
forms of bodily signalling. Rather, the most effective treatments However, social pain can also activate the sensorimotor system,
are aimed at processes of emotional regulation and reappraisal suggesting that ‘high level personal representations can entrain
that modulate activity in the AI-ACC. sensory processing in the absence of eliciting noxious stimulus’
Thus, a modular view of pain processing can point to evidence (Eisenberger 2012). These cases suggest a complex, continuous
of selective activation and independent regulation of sensorimo- structure to pain processing whose elements can be co-ordinated
tor and affective components of a pain processing system. The in a context-sensitive way.
 Pain suffering and the self. 5

We can add to this that most of the processing of pain signals dACC and AI may have a mediating role in the links between
is not performed by circuits specialized for responding to noxious social rejection and both inflammatory activity and depression
stimuli. Aversive responses to pain are sensori and viscero-motor. (Eisenberger 2012).
Attentional and inhibitory processes are amodal. For example,
dorsolateral prefrontal structures have an essential role to play This interpretation treats the AI as an integrative hub that func-
in pain modulation shown by the fact inhibition of activity in tions as a relay station between bodily regulation, emotional
these circuits removes the placebo response. These structures appraisal, and conceptual and narrative forms of self represen-
do not work in isolation but in co-operation with ventrolateral tation. Eisenberger is suggesting that this intermediary role for
and orbitofrontal structures involved in cognitive reappraisal and the AI goes both ways. Not only does the AI transcribe intero-
modulation of affective response and sensorimotor responses. ceptive and nociceptive experience to integrate bodily regulation
‘The descending modulatory systems involve brain regions that adaptively with higher level processing, it also transcribes high-
are important not only for pain but also for cognitive and emo- level personal and social and personal information (for example
tional functioning in general’(Bushnell et al. 2013). In other words, about ostracism or rejection) into formats that allow for adaptive
the regulatory role for dorsolateral circuitry is not restricted to low-level bodily responses. On one way of reading Eisenberger, the
pain modulation. nociceptive signal can be processed bottom up from sensorimotor
Similarly, empathetic responses, both affective and sensori- to conceptual-social levels or top down from conceptual-social to
motor, depend on activity in systems that one might think of sensorimotor. In both cases, the AI is an intermediate hub or relay
as, strictly speaking, not dedicated to processing noxious stimuli. station.
Empathy is strongly modulated by attachment and by the inter- The matrix conception suggests that pain processing activates
pretation of others mental states and attitudes, which in them- distributed circuitry across an essentially amodal matrix whose
selves have nothing to do with nociception. This is why the science elements include hubs of bodily/interoceptive, social emotional,
of pain has moved away from explaining pain experience as the and conceptual and executive processing. Activity across this
result of processing in a system specialized for pain processing matrix can be driven from any starting point. Anxious rumination
with discrete sensory and affective components. for example can make people prey to a range of distressing bodily
and emotional experiences, including the amplification of innocu-
ous nociceptive signals (Terasawa et al. 2013). Or, in prototypical
The pain matrix cases of pain, it can be driven by perception noxious stimuli such
The concept of a pain matrix was introduced by Melzack (1990) in as a burn or broken bone. In either case, the ultimate experience
the context of explaining the persistence of phantom limb pain ‘reflects the pattern of activity across the matrix’.
and its resistance to anaesthesia. He explained the persistence
of painful experience in the absence of body parts in terms of Nociceptive cortical processing is initiated in parallel in sen-
a distributed ‘neuromatrix’ that extends ‘through selective areas sory, motor, and limbic areas; it progresses rapidly to the
of the whole brain including the somatic, visual and limbic sys- recruitment of AI and fronto-parietal networks, and finally
tem’. The activation of elements of the matrix in the absence to the activation of perigenual, posterior cingulate and hip-
of sensory input can lead to exaggerated pain experience, not pocampal structures. Functional connectivity between sensory
least because there is no possibility of relieving a precipitating and high-level networks increases during the first second post-
bodily injury through movement or anaesthesia. Melzack argued stimulus, which may be determinant for access to conscious-
that the structure of the matrix is partly genetically specified (the ness (Garcia-Larrea and Bastuji 2018).
‘pain connectome’) but its final architecture is sensitive to devel-
opmental influences. The result is that the matrix can be activated The fact that there is a typical sequential pattern across the
via a variety of pathways depending on idiosyncratic patterns of matrix from peripheral (nociceptive) to more central processes
connectivity and conductivity. Chronic pain and ‘social’ pain for encourages the idea of an hierarchical modular structure to pain
example are evoked by different sources and sustained by a dif- processing if we concentrate on standard cases and neuropsy-
ferent pattern of activity across the matrix to pain directly evoked chological deficits. However, the matrix explains these cases in
by injury. The main point is that different types of pain experience terms of the role of hubs of cortical and subcortical processing
should be thought of as resulting from different ‘patterns of activ- whose activity can be initiated and maintained in different ways
ity across the whole network’ rather than as evidence of a modular in different contexts. The posterior insula is an integrative hub
architecture for pain (Melzack 1990). In particular, a strict double for basic bodily regulation of which nociceptive signals form an
dissociation between sensorimotor and affective processing is not important class. The role of the AI here is as a crucial integra-
supported by the evidence. tive relay station that allows bodily changes to be evaluated for
The matrix conception also undermines a strict bottom up or personal significance and social/personal information to entrain
feedforward model of pain processing in favour of one in which appropriate bodily responses. In order to play that role, the AI
pain experience is the emergent produce of recurrent processing communicates with hubs of circuits such as amygdala and ven-
across the matrix. ‘Social’ pain occasioned by ostracism or criti- tromedial prefrontal cortex that orchestrate emotional appraisal.
cism is an example. It is typically associated with activity in the As a recent review put it the insula is not an isolated ‘island’ but
AI circuitry which suggests an ‘affective’ interpretation: the sub- rather an integral brain hub connecting different functional sys-
ject is feeling personal distress. However, there are also cases in tems underlying sensory, emotional, motivational and cognitive
which social pain activates posterior systems involved in sensory processing (Gogolla 2017 585).
processing (Eisenberger 2012). One explanation is Activity across the insula reflect different ‘levels’ of integra-
tion with posterior insula a convergence zone for basic bodily
the role of the dACC and AI in responding to socially painful signals and the AI a relay station between posterior insula and
experience; these regions may be crucial for ‘translating’ expe- systems that determine the significance of those signals per-
riences of social disconnection into downstream physiological sonal/social goals. Partitioning of the insula is not sharp but
responses, which have implications for health. Indeed, the continuous.
6 Gerrans

The activation associated with both pain-related (posterior response selection, interoception, attention, response conflict,
insula) activation and that associated with prediction error autonomic arousal. Tellingly, functions non-specifically linked
(PE)-related (AI) activation correspond well with connectivity to pain are also carried out by the AI, which is implicated in
gradients observed along the posterior–anterior axis (Horing cognitive, affective, and regulatory functions, including intero-
et al. 2022). ceptive awareness, affective response, empathic processes, and
uncertainty (Betti and Aglioti 2016, 198).

This matrix conception also helps explain how it is that the

Activity in the AI provides a constant ‘locus of concern’ ensuring
Default Mode Network (DMN) can play a crucial role in modulat-
that activity across the brain is integrated for the organism’s ben-
ing the nociceptive signal. The DMN is a hub of autobiographical
efit. This does not mean that the AI is a ‘self module’, just that the
representation and crucial substrate of the narrative self (Davey
feeling of being a unified self is an emergent property of the inte-
et al. 2016). As such, within a broadly modular framework, one
gration of distributed activity to serve the goals of the organism
might regard its functioning as independent of sensory processing.
and that integration represents the system as a unified entity. This
But, in fact, it plays an important role in pain regulation largely
view is more consistent with the allostatic active inference view of
through its interactions with attentional and modulatory systems.
cognition which treats the mind as actively engaged in finding a
pattern of neural activity that optimizes functioning of the organ-
Activation in the salience network was found when attention ism. The self is inferred by the mind to be ‘the entity who benefits’
spontaneously focused on pain (20). In contrast, the DMN was when predictions are realized. When the AI is not active, sensation
engaged when attention was focused away from pain (20). ‘Indi- and cognition are not impaired but activity across the matrix is not
viduals’ intrinsic attention to pain’ (defined by the test–retest integrated to serve the goals of the subject. The resultant experi-
reproducibility of an individual’s tendency to attend away from ence is reported with a sense of detachment or depersonalization.
pain) was related to their structural and functional connec- Allostatic active inference has lost its centre of gravity.
tivity between DMN and the descending pain control system
(and the PAG in particular) (18). Also, alterations in the inter-
Asymbolia as depersonalization
play between the salience network, DMN, and descending pain
control network have been related to heightened attention to This framework supports the argument of Colin Klein against
pain in chronic pain patients (Wiech and Tracey 2013). deaffectualization theories of pain asymbolia. He suggested that
asymbolia is a species of ‘depersonalization for pain’. As he says

The ‘pain’ matrix and the role of self They recognise it as pain, but in some important sense it has
modelling ceased to be something worth caring about. It thus has the feel
When Melzack introduced the idea of a matrix, he treated it as the of a sensation which they can no longer identify with as their own
basis of a bodily sense of self. On his view, the matrix explains the (Klein 2015).
fact that
In this respect, his account recalls the explanation of classic deper-
sonalization (DPD) experience by (Michal et al. 2014) as ‘difficulties
experience of the body has a unitary, integrated quality that
of DPD patients to integrate their visceral and bodily perceptions
includes the quality of the ‘self’, the feeling that all the parts of
into a sense of their selves.’ In DPD, the inability to model experience
the body are uniquely one’s own (Melzack 1990).
as belonging to the self is global so that patients will report feeling
detached from ‘all’ experiences including pain. In pain asymbolia
Melzack was explaining why a person with a phantom limb that detachment is reported as applying specifically to pain. So the
still experiences that limb ‘as hers’. In order to act to avoid or ‘bodily perception’ not integrated into a sense of self is the noci-
reduce damage, the brain needs to treat experience produced by ceptive signal. This is why Klein (2015) argues that asymbolia is a
the matrix as a property of a unified persisting entity: the source form of depersonalization restricted to pain. As he puts it:
of bodily experience and target of regulation (Metzinger 2003; Ger-
rans 2014; Menon and Uddin 201; Limanowski and Blankenburg the phenomenology of asymbolia might resemble a kind
2013; Hohwy and Michael 2017; Seth and Tsakiris 2018). In the of depersonalization syndrome. … The asymbolic, and the
case of phantom limb, the model fails to update after amputation. depersonalized more generally, feel sensations that they are
The matrix idea fits with the view that the basic process- estranged from—that they do not take to be theirs in the sense
ing properties of the brain depend on the synchronized activity that we normally do. … (Klein 2015).
of large-scale networks (Pessoa 2017). These networks comprise
(at least) the salience network (allocation of cognitive resources This view, while attractive, is initially hard to square with neu-
whose top level is the AI-ACC circuit) the default mode network ropsychiatric theories of depersonalization that treat it as a form
(explicit episodic self-referential processing) and an executive net- of ‘dissociative deaffectualization’. For these theories, deperson-
work (high-level cognitive control) (Menon and Uddin 2010; Ger- alization results from involuntary inhibition of the AI as a disso-
rans 2014). Each of these networks has a proprietary network ciative response to intractable adversity such as trauma or abuse
architecture (Betti and Aglioti 2016). When we apply this idea to (Sierra 2008, 2009, Medford and Critchley 2010, Medford et al.
the processing of pain, the idea that there is a ‘sensorimotor cir- 2016). In other words, depersonalization theorists tend to endorse
cuit’ for initial processing of processing noxious stimuli and a a version of the modular theory of affective processing and treat
discrete ‘affective circuit’ centred on the AI-ACC dissolves. Instead, depersonalization as a loss of affect resulting from hypoactivity
the insula functions as a relay station between the salience sys- in the AI. This strategy exploits the brain’s natural opioid sys-
tem, sensorimotor processing, and the default mode network to tem (Sierra 2008). Morphine analgesia has the same dissociative
help configure activity in the matrix. This is consistent with the effect and seems to initially exploit the brain’s distress regulation
polymodal involvement of the insula in systems (Lee et al. 2014, Rütgen et al. 2015)
 Pain suffering and the self. 7

by initially targeting the AI. The functional magnestic reso- One difference between Craig’s account and mine is that Craig’s
nance imaging (FMRI) data suggest that opioid analgesics can account almost treats the AI as a discrete ‘self module’. My
directly influence emotional responses at low doses that do not view is the more modest one that the AI is an integrative hub
alter sensory aspects of pain (Lee et al. 2014). whose activity transcribes the interoceptive signal into one that
signals emotional salience. The emergent result of this integra-
tive process is the experience of being the subject of salient
In general, depersonalized patients who are deaffectualized have
experience.
intact basic bodily awareness and regulation. A reasonable infer-
To return to the case of pain asymbolia, in pain asymbolia, dys-
ence is that their posterior insula functioning is unimpaired but
function of posterior insula leads to failure to model nociceptive
they feel depersonalized in virtue of hypoactivity in the AI. That is
signals as belonging to the bodily self. This produces the experi-
the deaffectualization theory of depersonalization in a nutshell.
ence of sensations of bodily damage (or other forms of threat to
Recall however that Klein proposed the depersonalization theory
bodily integrity) that do not belong to ‘me’. This failure to incorpo-
of asymbolia as an ‘alternative’ to the idea that it is a form of deaf-
rate nociceptive signals at the level of bodily self-modelling means
fectualization for pain. We should also recall that asymbolia is
that they do not entrain regulatory responses, including higher
agreed on all sides to be the result of damage to posterior insula,
levels of active inference that depend on AI activity. If the sig-
hypothesized to be an integrative hub for basic bodily regulation
nals are not sensed as belonging to me at the most basic level
not affected.
of selfhood, there is no need to establish their emotional salience.
Consequently, they are not affectively transcribed. So, they do not
Reconciliation. The self model lead to the feelings of personal distress and attempts to reduce
them. The narrative I report is the sensation of nociceptive expe-
These facts are hard to reconcile prima facie. However, the role rience that is not felt to matter to the self. Pain aysmbolia is thus a
of the insula in self modelling explains them. The posterior case of processing nociceptive signals ‘outside’ the self-modelling
insula cortex integrates values of basic bodily variables like blood hierarchy. That is to say without those signals being attributed to
pressure and hydration as well as nociception (bodily damage) the self whose goal structure determines the regulatory response
to co-ordinate basic regulatory functions. The posterior insula to sensory experience. This explains why, as Klein (2015) puts it,
is thus the primary substrate of interoceptive experience and asymbolia is a kind of ‘[I]ndifference. One’s body becomes, as it
hence of what Anil Seth called ‘material me’, the feeling of being were, just another object in the world’.
an embodied self. The AI cortex integrates interoceptive signals
from the posterior insula with information from other channels.
This enables bodily signals (such as nociception) to be contex- Acknowledgements
tualized and managed adaptively by entraining the full range Research for this paper was supported by the Australian Research
of regulatory capacities. The result is that we do not just feel Council. Discovery Grant DP180101323.
like the subject of bodily states but of emotional states that This is a theoretical paper and data are not acquired in its prepa-
reflect our goals. In a dangerous episode, we do not just feel ration.
adaptive bodily responses but we feel them as fearful because
those bodily responses are produced as ways of realizing a goal
of avoiding danger.This is why the concept of allostatic active Conflict of interest
inference helps explain why the pain matrix, though essentially
None declared.
amodal, can give rise to such various phenomenology. Allostatic
active inference is a process of recruiting and co-ordinating a
suite of systems to maintain systemic integrity. In the case of Data availability
pain this means detecting, avoiding, responding to, and repair-
No new data were generated or analysed in support of this
ing damage using relevant resources. This requires integrating
research.
relevant systems. At the most basic level, managed by the poste-
rior insula, this integration is felt as intero/nociceptive experience
and embodied selfhood. The AI relays the interoceptive signal References
to other systems that help interpret and manage it adaptively.
Barrett LF. The theory of constructed emotion: an active inference
This transcribes the interoceptive signal allowing it to be expe-
account of interoception and categorization. Soc Cognit Affective
rienced as an affect. For example, an abdominal cramp will be
Neurosci 2017;12:1–23.
felt quite differently if it is experienced transiently in the gym
Barrett LF, Quigley KS, Hamilton P et al. An active inference theory
doing sit-ups or if it is experienced by someone who is fearing a
of allostasis and interoception in depression. Philos Trans R Soc B
miscarriage.
2016;371:20160011.
Thus, it is not surprising to see that contemporary affective
Berthier M, Starkstein S, Leiguarda R et al. Asymbolia for pain: a
neuroscience treats experience produced by activation of the AI as
sensory-limbic disconnection syndrome. Ann Neurol 1988;24:41–9.
a form of higher order bodily representation that represents the
Betti V, Aglioti SM. Dynamic construction of the neural net-
integrated functioning of the organism ‘evaluated against emo-
works underpinning empathy for pain. Neurosci Biobehav Rev
tionally salient goals creating a sense of self in the process’. As
2016;63:191–206.
Bud Craig (2009) puts it,
Billon A. Making sense of the Cotard syndrome: insights from the
study of depersonalisation. Mind Lang 2016;31:356–91.
the integration successively includes homeostatic, environ- Billon A. What is it like to lack mineness? Depersonalization as
mental, hedonic, motivational, social, and cognitive activity to a probe for the scope, nature, and role of mineness. In: Guil-
produce a ‘global emotional moment’, ‘which represents the lot M, Garcia-Carpintero M (eds.), Self-Experience: Essays on Inner
sentient self at one moment of time’ (Craig 2009). Awareness. Oxford: OUP, 2023, 314–42.
8 Gerrans

Bushnell MC, Ceko ̌ M, Low LA et al. Cognitive and emotional con- Medford N, Critchley HD. Conjoint activity of anterior insular and
trol of pain and its disruption in chronic pain. Nat Rev Neurosci anterior cingulate cortex: awareness and response. Brain Struct
2013;14:502–11. Funct 2010;214:535–49.
Corcoran AW, Hohwy J. 2017. Allostasis, interoception, and the free Medford N, Sierra M, Stringaris A et al. Emotional experience and
energy principle: feeling our way forward. awareness of self: functional MRI studies of depersonalization
Craig AD. How do you feel—now? The anterior insula and human disorder. Front Psychol 2016;7:432.
awareness. Nat Rev Neurosci 2009;10:59–70. Melzack R. Phantom limbs and the concept of a neuromatrix. Trends
Danziger N. Bases neurologiques de l’affect douloureux. Rev Neurol Neurosci 1990;13:88–92.
2006;162:395–9. Menon V, Uddin LQ. Saliency, switching, attention and control: a net-
Davey CG, Pujol J, Harrison BJ. Mapping the self in the brain’s default work model of insula function. Brain Struct Funct 2010;214:655–67.
mode network. Neuroimage 2016;132:390–7. Menon V and Uddin LQ. Saliency, switching, attention and con-
Duquette M, Roy M, Leporé F, Peretz I, Rainville P. Mécanismes trol: a network model of insula function. Brain Struct Funct
cérébraux impliqués dans l’interaction entre la douleur et les 2010;214:655–67.
émotions. Revue Neurologique 2007;163:169–79. Mesulam -M-M. Principles of Behavioral and Cognitive Neurology. Oxford:
Eisenberger NI. The pain of social disconnection: examining the Oxford University Press, 2000.
shared neural underpinnings of physical and social pain. Nat Rev Metzinger T. Being no one. Cambridge Mass: MIT Press, 2003.
Neurosci 2012;13:421–34. Michal M, Reuchlein B, Adler J et al. Striking Discrepancy of
Fallon N, Roberts C, Stancak A. Shared and distinct functional net- Anomalous Body Experiences with Normal Interoceptive Accu-
works for empathy and pain processing: a systematic review racy in Depersonalization-Derealization Disorder. PLoS ONE
and meta-analysis of fMRI studies. Social Cognitive and Affective 2014;9:e89823.
Neuroscience 2020;15:709–23. Moayedi M. All roads lead to the insula. Pain 2014;155:1920–1.
Ferraro S, Klugah-Brown B, Tench CR et al. Dysregulated ante- Moutoussis M, Fearon P, El-Deredy W et al. Bayesian inferences about
rior insula reactivity as robust functional biomarker for chronic the self (and others): a review. Conscious Cogn 2014;25:67–76.
pain—meta-analytic evidence from neuroimaging studies. Human Pessoa L. A network model of the emotional brain. Trends Cogn Sci
Brain Mapp 2022;43:998–1010. 2017;21:357–71.
Frot M, Faillenot I, Mauguière F et al. Processing of nociceptive input Rütgen M, Seidel E-M, Silani G et al. Placebo analgesia and its opi-
from posterior to anterior insula in humans. Human Brain Mapp oidergic regulation suggest that empathy for pain is grounded in
2014;35:5486–99. self pain. Proc Natl Acad Sci 2015;112:E5638–46.
Garcia-Larrea L, Bastuji H. Pain and consciousness. Prog Neuro Psy- Schilder P, Stengel E. Asymbolia for pain. Archives of Neurology &
chopharmacol Biol Psychiatry 2018;87:193–9. Psychiatry 1931;2:598–600.
Gerrans P. The Measure of Madness Philosophy and Cognitive Neuropsychi- Seth AK. Interoceptive inference, emotion, and the embodied self.
atry. Cambridge Mass: MIT Press, 2014. Trends in Cognitive Sciences 2013;17:565–73.
Gerrans P. Pain asymbolia as depersonalization for pain experi- Seth AK, Tsakiris M. Being a beast machine: the somatic basis of
ence. An interoceptive active inference account. Front Psychol selfhood. Trends Cogn Sci 2018;22:969–81.
2020;11:523710. Sierra M. Depersonalization disorder: pharmacological approaches.
Gogolla N. The insular cortex. Curr Biol 2017;27:R580–6. Expert Rev Neurotherapeutics 2008;8:19–26.
Grahek N. Feeling Pain and Being in Pain. Cambridge Mass: MIT Press, Sierra M. Depersonalization: A New Look at A Neglected Syndrome. Cam-
2011. bridge: Cambridge University Press, 2009.
Gu X, Hof PR, Friston KJ et al. Anterior insular cortex and emotional Sierra M, Senior C, Dalton J et al. Autonomic response in depersonal-
awareness. J Comp Neurol 2013;521:3371–88. ization disorder. Arch Gen Psychiatry 2002;59:833–8.
Hohwy J, Michael J. 16 Why should any body have a self? Simons LE et al. Psychological processing in chronic pain: a neural
In: De Vignement F, Alsmith A (eds.), The Subject’s Matter systems approach. Neurosci Biobehav Rev 2014;39:61–78.
Self Consciousness and the Body. Cambridge Mass: MIT, 2017, Singer T, Lamm C The social neuroscience of empathy Annals of the
363–71. New York Academy of Sciences 2009 1156 81–96
Horing B, Büchel C, Summerfield C. The human insula processes both Singer T, Seymour B, O’Doherty J et al. Empathy for pain involves
modality-independent and pain-selective learning signals. PLoS the affective but not sensory components of pain. Science
Biol 2022;20:e3001540. 2004;303:1157–62.
Kleckner IR, Zhang J, Touroutoglou A, et al. Evidence for a large-scale Starr CJ, Sawaki L, Wittenberg GF et al. Roles of the insular cortex
brain system supporting allostasis and interoception in humans. in the modulation of pain: insights from brain lesions. J Neurosci
Nat Human Behav 2017;1:0069. 2009;29:2684–94.
Klein C. What pain asymbolia really shows. Mind 2015;124: Terasawa Y, Shibata M, Moriguchi Y et al. Anterior insular cortex
493–516. mediates bodily sensibility and social anxiety. Soc Cognit Affective
Lee MC, Wanigasekera V, Tracey I et al. Imaging opioid analge- Neurosci 2013;8:259–66.
sia in the human brain and its potential relevance for under- Tschantz A, Barca L, Maisto D et al. Simulating homeostatic, allostatic
standing opioid use in chronic pain. Neuropharmacology 2014;84: and goal-directed forms of interoceptive control using active
123–30. inference. Biological Psychology 2022;169:108266.
Limanowski J, Blankenburg F. Minimal self-models and the free Wang X, Tan Y, Van den Bergh O, von Leupoldt A, Qiu J. Intrinsic
energy principle. Front. Hum. Neurosci. 2013;7. functional brain connectivity patterns underlying enhanced inte-
Li Y, Zhang T, Li W et al Linking brain structure and activation in ante- roceptive sensibility. Journal of Affective Disorders 2020;276:804–14.
rior insula cortex to explain the trait empathy for pain Human Wiech K, Tracey I. Pain, decisions, and actions: a motivational per-
brain mapping 2020 41 1030–42 spective. Front Neurosci 2013;7:46.
 Pain suffering and the self. 9

Xu A. Convergent neural representations of experimentally-induced Yarns BC, Cassidy JT, Jimenez AM et al. At the intersection of anger,
acute pain in healthy volunteers: A large-scale fMRI meta- chronic pain, and the brain: a mini-review. Neurosci Biobehav Rev
analysis. Neuroscience & Biobehavioral Reviews 2020;112:300–23. 2022;135:104558.
Yao S, Becker B, Zhao W et al. Oxytocin Modulates Attention Switching Zhou F, Li J, Zhao W et al. Empathic pain evoked by sensory
Between Interoceptive Signals and External Social Cues. Neuropsy- and emotional-communicative cues share common and process-
chopharmacol. 2018;43:294–301. specific neural representations. eLife 2020;9.
Neuroscience of Consciousness, 2024, 2024(1), 1–9
DOI: https://doi.org/10.1093/nc/niae002
Research Article
Received 31 May 2023; Revised 6 January 2024; Accepted 16 January 2024
© The Author(s) 2024. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.