ABSTRACT
Pain and suffering often bundle damage, nociception, unpleasantness, memory, control, temporal persistence, and phenomenal character into one explanatory target. This paper develops a progressive attribution method that separates these dimensions through controlled conceptual variation. Beginning from paradigmatic first-person pain, it varies perspective, present harm, memory, control, biological architecture, and physical realization before considering an independently artificial experiential candidate. The sequence yields two central results. First, architectural variation and substrate variation are distinct attribution problems. Second, both preserved and changed attribution incur a comparative relevance burden: similarity and difference require independent reasons for treating particular properties as experience-relevant. Reconstructing aversive states as interacting regulatory, affective, temporal, and self-relevant processes clarifies their organization without settling phenomenal aversion. Progressive decomposition therefore does not solve why a state is bad to undergo; it isolates the commitments required to explain and attribute that property.
KEYWORDS
pain; suffering; aversive experience; consciousness; temporality; animal sentience; artificial consciousness
Word count: 8,467 (including title, abstract, keywords, headings, references, and AI-use disclosure; excluding this count statement).
1. Introduction and Method
Begin with a case that seems to require almost no theory.
Something hurts.
From the first-person position, the occurrence of the experience does not normally wait for a definition of pain, a neural measurement, or an argument about consciousness. One may be mistaken about the cause. One may misjudge the severity, duration, or bodily source. One may possess a false theory of what pain is. Yet the aversive occurrence itself presents differently from an inference about somebody else.
This paper begins there not because first-person experience settles the metaphysics of pain, but because it supplies the paradigmatic case from which the later attribution problem arises.
Contemporary pain science already distinguishes pain from nociception and warns against inferring pain from sensory-neural activity alone (Raja et al. 2020). That is enough for the opening point: tissue damage, nociception, and pain should not be treated as identical.
Philosophy complicates the category further. What belongs to pain itself? Sensory quality? Unpleasantness? Motivation? Attention? Bodily representation? Action tendency? The broader disruption pain causes in thought and behavior? Different theories draw the boundaries differently, and unusual clinical cases make the boundaries harder rather than easier to locate (Coninx 2024).
Starting by defining pain too tightly would therefore risk deciding the argument before it begins.
Instead, the paper will proceed by controlled variation. Begin with the paradigmatic first-person case and alter properties ordinarily associated with it: perspective, actual bodily harm, memory, temporal integration, control, nervous architecture, and eventually physical realization. At each transition the cases incur a comparative relevance burden: if attribution changes, what differing property independently warrants that change; if attribution is preserved despite architectural or physical difference, what preserved property warrants invariance? Candidate relevance can be supported by causal intervention or systematic covariation in accepted experiential cases, by necessity within a defended theory, or by explanatory discrimination between otherwise similar cases. None of these is by itself a proof of phenomenology.
The question is intentionally narrower than “What is consciousness?” and broader than “Which neurons encode nociception?” It concerns the conditions under which we attribute something that is bad to undergo.
For convenience, the paper will sometimes use “consciousness” to refer to the presence of subjective experience. Nothing in that usage commits the argument to consciousness as a separate substance, a fundamental property, a functional organization, an illusion, or any other contested ontology.
Likewise, “aversive experience” is not introduced as a solved technical category. It marks the target we are trying to understand while ordinary terms such as pain and suffering are progressively decomposed.
The contribution is therefore primarily diagnostic and methodological rather than a new unitary theory of pain. The method will not produce certainty at every stage. Its purpose is to discover where certainty is lost, what evidence replaces it, which differences actually do explanatory work, and where competing theories place their remaining explanatory burden.
Suppose the pain is mine.
From here, I do not ordinarily infer that it hurts from my withdrawal, heart rate, neural activity, or verbal report. Those can tell me things about the state, but they are not why its occurrence is evident to me. The experience is presented first-personally.
Now perform a conceptual operation.
Observe the same event as though the person in pain were somebody else.
The observer sees a stimulus, perhaps tissue disturbance, nociceptive activity, changes in attention, autonomic responses, protective movement, learning, verbal report, and attempts to terminate the state. With sufficient technology, the observer might obtain an extraordinarily detailed description of neural and bodily activity.
Where, in that description, is the hurting?
Call this the Third-Person Self Test.
The question is not asking for a spatial location where a mysterious substance should be found. Nor does failure to find a separate “painfulness signal” show that painfulness is nonphysical. The point is epistemic. The underlying candidate event has not been changed. Only the route by which it is known has changed.
From the first-person position, “this hurts” functions as immediate evidence of the experience. From the third-person position, the evidence becomes physiological, behavioral, functional, contextual, and report-based.
Now replace the observed person with another human.
The epistemic structure is suddenly familiar. I do not experience another person’s pain. I infer it. The inference is usually extraordinarily strong because humans share anatomy, nervous systems, development, behavior, language, and a vast causal background. If another person burns a hand, recoils, protects it, describes intense pain, and exhibits the expected physiological responses, radical skepticism about whether anything hurts is normally unreasonable.
But the experience itself has not become directly observable.
This is the classical problem of other minds in a particularly concrete form. The paper will not solve it. What matters is that nearly every attribution of pain beyond one’s own case already depends on indirect evidence.
That observation will become increasingly important as similarity declines.
Another human provides overwhelming analogical evidence. A nonhuman animal may share less. A gradually synthetic human changes the physical realization. An independently artificial system may share still less of the biological background.
At each stage the temptation will be to convert difference into absence.
The Third-Person Self Test gives us a constraint against doing so too quickly. If the third-person description of my own pain contains only neural, bodily, behavioral, representational, and functional facts, then the mere fact that another system is also described in those terms cannot by itself show that its state lacks phenomenology. The same descriptive limitation applies to the one case whose phenomenology was stipulated at the beginning.
This does not establish symmetry among all other minds. Evidence matters. Another human and an artificial system may have radically different evidential support. Shared biology can be relevant. Architecture can be relevant. Developmental history can be relevant. The point is only that absence of direct phenomenal access is not a special defect that begins when the subject becomes unfamiliar.
We therefore need to know what evidence we are actually tracking.
The method therefore begins by opening the ordinary concept of pain before moving farther from the human case.
AI assistance in manuscript preparation
OpenAI ChatGPT (GPT-5.6 Sol, GPT-6 Astra, and GPT-6 Sol; used during 2026) was used as an AI-assisted research and writing tool throughout development of this manuscript, including literature discovery, source verification, argument stress-testing, structural development, drafting, rewriting, and bibliographic checking. The author originated the paper’s core ideas, research questions, and principal thought experiments, directed the analysis and revisions, reviewed the manuscript in full, verified the cited sources, and accepts full responsibility for the accuracy, originality, and conclusions of the work.
2. Decomposing Pain and Representation
Once pain is viewed from the outside, decomposition is almost irresistible. Noxious stimulation can be distinguished from neural processing; sensory intensity from unpleasantness; unpleasantness from motivation; motivation from action; attention from memory; immediate response from the broader effects that pain has on a person’s mental life. The ordinary sentence “this hurts” appears to conceal a system of distinguishable processes.
The IASP definition likewise treats pain as an unpleasant sensory and emotional experience and explicitly distinguishes it from nociception (Raja et al. 2020). Whatever philosophical account one accepts, pain is therefore not scientifically identified with a noxious stimulus or sensory-neural activity alone.
Pain is commonly analyzed in sensory-discriminative, affective-motivational, and cognitive-evaluative terms, distinctions useful because features of an episode can vary and respond differently to intervention (Coninx 2022; Fulkerson 2026).
Analytical distinction should not be mistaken for mechanistic independence. Experimental attempts to dissociate pain intensity from unpleasantness provide suggestive but contested evidence: hypnotic-suggestion studies reported partial dissociations, while a later systematic review found no compelling evidence that the dimensions could be reliably manipulated independently (Rainville et al. 1999; Talbot et al. 2019). Pain asymbolia sharpens the same problem. Cases traditionally interpreted as preserved sensory pain with diminished affective response have also been interpreted as showing that the residual state should no longer count as pain at all (Griffith and Kind 2024; Duval and Klein 2026). The lesson needed here is methodological: a dissociation does not interpret itself. Observing features come apart does not yet tell us whether we have separated components of one phenomenon, changed the phenomenon’s kind, or crossed a classificatory boundary.
Decomposition creates a second problem: once sensory, affective, evaluative, motivational, and cognitive dimensions are distinguished, nothing about the distinction itself tells us which belong to pain proper, where pain ends, or whether the dimensions correspond to independent mechanisms. Existing accounts disagree sharply about those questions, ranging from eliminativist and family-resemblance approaches to multidimensional and psychological-integration models (Corns 2016; Coninx 2022, 2023, 2024; Fulkerson 2026).
For the present argument, that disagreement motivates a narrower methodological rule: analytical separability does not imply ontological separateness. Dimensions can be useful for explanation and intervention without corresponding to independent modules or sharp natural borders. Graded transitions—discomfort into pain, concern into fear, frustration into distress—reinforce the caution: boundaries may exist, but they must be demonstrated rather than inherited from the nouns we use. Conversely, interaction among dimensions does not make the distinctions useless; coordinated processes can still form stable and scientifically informative patterns.
For the argument that follows, this gives decomposition a specific role. Pain remains the paradigmatic starting case because it provides the strongest first-person and empirical anchor, but the target now broadens to aversive experience where the relevant dimensions extend beyond nociceptive pain. This does not classify grief, frustration, or later artificial damage-related states as pains. We will use dimensions as intervention coordinates. We will ask what happens when actual bodily harm is removed, when memory and temporal integration are altered, when control changes, when nervous architecture changes, and eventually when physical substrate changes. In each case, the conceptual intervention will try to vary one relation while holding others sufficiently stable to expose what a theory treats as relevant.
These interventions are idealizations. In an actual organism, changing memory may alter attention; changing control may alter prediction; changing sensory processing may alter motivation. The point of an idealized intervention is not to deny those interactions. It is to make theoretical commitments visible. Empirical cases can then tell us how far the dimensions can in fact be varied independently.
Pain is therefore being taken apart here without any promise that the resulting pieces correspond to nature’s own seams. We may discover family resemblances, integrated kinds, graded boundaries, context-sensitive clusters, or something else. What matters at this stage is that the ordinary unity of “pain” no longer gets to decide the result in advance.
Representation Without Present Harm
Pain is strongly associated with injury, making bodily condition and experienced state easy to fuse conceptually.
But the state of the body and the state represented in experience are not the same variable.
Dreams provide a compact demonstration.
Dream reports provide the needed dissociation. Pain can occur within a dream world whose represented injury and context do not match the waking body; laboratory cases sometimes incorporate real somatosensory stimulation, while ordinary dream reports can include painful injuries that never occurred externally (Nielsen et al. 1993; Zadra et al. 1998). The case does not show that bodily input is never involved, nor does it challenge representational theories of pain. It establishes only what the present intervention needs: actual present tissue damage and experienced or represented pain are not identical variables.
One assumption has therefore been loosened. Once absent external states can participate in present experience, another question follows naturally. Past events no longer exist, and future events do not yet exist. Yet memory and anticipation can make both matter now.
3. Time, Memory, and Control
Suffering appears to extend through time. A person who has been in pain for six hours does not ordinarily describe six hours as a mere collection of unrelated painful instants. The present can contain the past—“this has been happening for hours”—and it can contain the future—“I may have six more hours of this.” Memory, anticipation, perceived persistence, and the evaluation of an episode after it ends can therefore become entangled with whatever is occurring in the present moment. Yet these temporal relations are easy to collapse into a single question: how long did the suffering last?
That question is less simple than it appears. Objective duration, felt duration, memory for an episode, memory for the qualitative character of pain, representation of persistence, anticipation of recurrence, retrospective evaluation, and welfare across time are not interchangeable concepts. This section separates them. The purpose is not to argue that memory is necessary for aversion, nor that one theory of welfare aggregation is correct. It is to ask what changes when temporal access itself is varied.
Several empirical and philosophical results already separate capacities that ordinary talk of “remembering pain” tends to bundle together. Affective states can persist after explicit memory for their eliciting events is severely impaired; remembering an event, its cause or intensity, retaining its affective consequences, and integrating it into an autobiographical history are distinct capacities; and retrospective evaluation need not reproduce the moment-by-moment sequence that occurred (Feinstein, Duff, and Tranel 2010; Guzmán-Vélez, Feinstein, and Tranel 2014; Montero 2020; Coninx 2020; Fredrickson and Kahneman 1993; Kahneman et al. 1993). None of this shows that memory is irrelevant to suffering. It shows that “memory” cannot be treated as one variable.
Before any thought experiment, six temporal questions should be distinguished: current aversive organization; subjective or felt duration; cross-temporal integration (including episodic recall, future simulation, psychological connectedness, and representation of persistence); retrospective evaluation; historical aversive exposure; and diachronic welfare. The last is a normative question about welfare across an interval, not merely a description of what aversive states occurred.
Consider a Memory Gradient. Begin with an ordinary person experiencing an aversive state while possessing normal autobiographical memory and anticipation. Reduce access to the preceding history: an hour, ten minutes, one minute, ten seconds, one second. At the limiting edge, imagine extremely weak cross-episode integration while preserving whatever organization is stipulated to support immediate aversion.
The gradient is diagnostic, not a prediction that suffering decreases smoothly with memory. It asks where a theory changes its attribution of current aversion, welfare, or suffering—and what property justifies either a threshold or a gradual change.
The limiting case makes the problem vivid. Suppose a system undergoes an intensely aversive experiential moment. Before the next episode, the information required to represent the previous episode as part of its own continuing experiential history becomes inaccessible. Another intensely aversive moment occurs. Again cross-episode access is lost. Imagine this continuing for ten years.
Call this the Ten-Year Second.
The case is a controlled version of the familiar forgotten-pain problem: it holds immediate aversion high while driving one form of cross-temporal integration toward zero, without building a verdict about welfare into the description.
The case separates questions that ordinarily travel together. Each moment is stipulated to be aversive, but no moment represents years of persistence. An extensive aversive history occurs, yet whether it constitutes ten years of diachronic suffering in the ordinary sense does not follow without a theory of welfare and temporal integration.
An additive answer treats the many aversive moments as correspondingly large suffering; a continuity-sensitive answer may reduce the welfare significance because persistence is never represented. The thought experiment does not choose between them. It exposes what each answer requires.
The welfare question cannot be settled by temporal integration alone. Felt duration need not straightforwardly determine welfare contribution, while severed psychological continuity may itself matter prudentially (Mogensen 2026; Rebello 2026). Even if weak connectedness complicates whether the sequence belongs to one persisting psychological subject, it does not erase the aversive character stipulated within each moment. Personal identity, temporal integration, historical exposure, and welfare aggregation therefore remain distinct questions.
Two inversions complete the separation. A false autobiographical memory can generate present distress without creating the aversive history it represents; prolonged anticipation can generate suffering even when the feared event never occurs. Together with the Ten-Year Second, these cases separate historical exposure, temporal integration, retrospective representation, and future-directed suffering.
Mendl and Paul (2008) extend the distinction to animal welfare: limited episodic mental time travel may constrain rumination or future anxiety without eliminating affective carryover through learning, stress, or mood. Temporal integration is therefore one dimension of suffering, not a universal meter of sentience.
This becomes important later: systems can differ dramatically in temporal integration without licensing the inference that weaker autobiographical continuity means absence of immediate aversion.
The lesson of this section is therefore deliberately modest. Suffering through time is not one variable. Current aversive organization, felt duration, cross-temporal integration, retrospective evaluation, historical exposure, and diachronic welfare can come apart conceptually and, in limited cases, empirically. How they interact is a further question. Whether welfare across time should be additive, holistic, sequence-sensitive, identity-relative, or something else is not settled here.
What has been established is a constraint on the larger investigation: when memory, temporal integration, or anticipation changes, we cannot simply announce that “suffering” has increased, decreased, appeared, or disappeared. We must specify which temporal relation changed and why that change is relevant to the particular attribution being made.
Control, Preference, and Persistence
Imagine watching a film you enjoy.
Watch it again tomorrow.
Then again.
Now imagine being required to watch the same film three times every day for five years.
The images and sounds need not become physically damaging. Yet the situation could become intensely aversive. The subject wants another state, cannot obtain it, repeatedly encounters the same unwanted condition, and may come to anticipate its continuation.
Now alter one fact.
The door is unlocked. The film can be stopped at any moment.
The sensory content has not changed. The relation between the subject and the state has.
The Forced Movie is intentionally not an example of nociceptive pain. Its purpose is to expose dimensions of aversive organization that bodily injury can obscure: preference conflict, persistence, agency, escape, and perceived control.
Pain research provides a narrower empirical analogue. In an experiment on controllability, healthy participants received painful electrical stimulation under conditions they could terminate and conditions whose termination was attributed to a computer (Löffler et al. 2018). In the first sample, controllability reduced reported pain-related suffering without reducing pain intensity or unpleasantness. The effect depended on how personal control was framed and did not reproduce identically in a second sample with altered instructions; there, the pattern changed.
The correct conclusion is therefore not that control universally reduces suffering. It is that personal controllability can alter the experienced organization of a painful episode in ways not captured simply by stimulus intensity.
That result matters because several philosophical theories place different weight on this relation.
A desire-based account may emphasize wanting the state to stop. An imperativist may emphasize the directive structure: protect, escape, terminate. An appraisal theory may treat coping potential and controllability as dimensions that reorganize the emotional episode (Scherer 2009).
None of these interpretations should yet be declared correct.
The important distinction is between properties of the stimulus and properties of the subject’s relation to the state.
A painful electrical stimulus can remain physically similar while perceived control changes. A harmless film can become aversive through forced repetition and inability to exit. Conversely, a painful condition can sometimes become more tolerable when its end is predictable or controllable even though the nociceptive input remains.
Preference conflict alone cannot be identified with suffering. A system can contain unsatisfied goals without feeling anything. Nor can failed correction by itself be identified with pain. Those possibilities will become important in the artificial case.
But control, persistence, and attempted termination have now earned places in the analytical landscape. They are not decorative psychological reactions added after “the real pain.” In at least some circumstances they alter the broader state through which pain becomes suffering.
Having varied temporal and agential relations while keeping the subject human, the next intervention changes architecture.
4. Other Minds and Animal Architecture
The problem of other minds does not disappear when the other mind belongs to another species. If anything, the inferential distance increases. Another human shares much of one’s biology, development, behavior, language, and neural organization. A nonhuman animal may share some of those features and differ radically in others. Yet comparative science does not respond to that distance by requiring direct access to subjective experience before making any attribution at all; it works with converging evidence (Birch, Schnell, and Clayton 2020).
That practice matters for the present argument because architectural difference is often treated informally as if it were itself evidence against experience. The animal case shows why that inference is too fast.
Comparative attribution already operates across substantial architectural differences. Rather than placing species on a single human-centered consciousness ladder or treating one behavior as a sentience detector, it relies on converging indicators (Birch, Schnell, and Clayton 2020). Cephalopods provide a useful stress test: octopus evidence combines flexible avoidance, relief-seeking, site-directed protection, and sustained neural processing, while broader assessments reach graded conclusions across cephalopod groups rather than treating unfamiliar architecture as decisive (Crook 2021; Schnell et al. 2026). The inference remains third-person and defeasible, but it illustrates the relevant epistemic pattern: confidence can be graded as evidence converges even when architecture differs radically.
Imagine beginning with another human and progressively altering architecture. Some evidence that initially carries enormous weight becomes weaker. Shared language may disappear. Detailed neural homology may decline. Behavioral repertoires change. Other evidence may become comparatively more important: flexible learning, motivational tradeoffs, protective behavior, analgesic preference, integrative processing, and the way those indicators converge. Confidence can change without requiring a magical boundary at which experience must suddenly vanish.
The point is not that every architectural difference is irrelevant. Quite the opposite. The question is which differences are relevant and why.
A biological essentialist can accept all of this: comparative animals remain living products of evolution. The Animal Bridge therefore establishes only that exact human neural architecture is not required for serious affective attribution; it does not establish substrate independence.
Architecture and substrate must therefore remain distinct. An octopus varies organization dramatically relative to a human while remaining a biological nervous system; the next intervention varies physical realization more directly.
The earlier temporal result carries over directly: species may differ in episodic memory, future simulation, and integration across time without those differences functioning as a universal meter of sentience. Limited mental time travel may constrain rumination or anticipatory anxiety while leaving learning, stress, and mood effects intact (Mendl and Paul 2008).
This matters because a familiar mistake can now be avoided. If an organism cannot represent “I have been suffering for three days” in the way an adult human can, it does not follow that no aversive experience occurs. It may instead lack one particular temporal organization of suffering. Conversely, granting immediate aversion does not justify attributing every human form of dread, rumination, grief, or autobiographical distress to that organism. The framework allows difference without forcing absence.
Comparative evidence cannot eliminate the possibility of sophisticated nonphenomenal nociception. Its methodological lesson is narrower: as architecture changes, confidence should track evidence rather than remain constant or collapse automatically to zero. Difference matters when there is an independent reason to treat the difference as experience-relevant. The Animal Bridge therefore leaves biological essentialism intact while removing the easier assumption that departure from human architecture is sufficient by itself.
To test substrate rather than architecture, the next move should not be to jump from an octopus to a computer. It should begin where the paper began—with the human case—and change the physical realization gradually. If the organization relevant to pain were progressively preserved while biological components were replaced, when, if anywhere, should the attribution of aversive experience change?
5. Substrate and Artificial Systems
The Animal Bridge changed architecture while leaving one fact largely untouched: every system considered remained biological. A biological essentialist can therefore accept the entire preceding section. Humans, mammals, birds, and cephalopods may realize experience through very different nervous architectures, the objection goes, while still sharing some physical property of living nervous systems that is necessary for consciousness or aversion. The animal evidence does not tell us whether that is true.
Suppose a human subject is experiencing a paradigmatic aversive bodily pain. Replace one neuron participating in the relevant processing with an artificial component. The replacement preserves the neuron’s specified causal interactions with the rest of the system: it receives the same relevant inputs, produces the same relevant outputs, and participates in the same fine-grained organization at the level stipulated by the experiment. Then replace another. And another. Continue gradually until progressively larger portions of the relevant nervous system are artificially realized.
At what point, if any, should attribution of the aversive experience change?
The replacement case inherits the familiar dispute over multiple realization: preserving a specified causal or organizational description does not establish that every property relevant to consciousness has been preserved (Lewis 1980; Chalmers 1995; Polger and Shapiro 2016). “Functionally equivalent” therefore means equivalent only at the stipulated level of organization; a critic may still identify an omitted physical property as experience-relevant. This makes gradual replacement a test of relevance rather than a shortcut to functionalism. Chalmers’s replacement strategy supplies the closest precedent, but here the intervention is narrowed to aversive experience.
Call the case the Synthetic Pain Continuum.
Begin with the human subject in an aversive state. At the biological starting point, the subject reports that the state hurts, attempts to terminate it, directs attention toward it, protects the affected area, learns from it, and exhibits whatever other processes our developing account treats as relevant. As replacement proceeds, preserve the specified organization supporting those capacities. At every stage the subject continues to report the same aversive character.
One response is straightforwardly functionalist: if the relevant organization remains, so does the aversive state. On that view, the percentage of biological material is beside the point.
A substrate-sensitive theory can instead hold that replacement removes experience-relevant causal properties of living neural tissue—biochemical, metabolic, electromagnetic, multiscale, or otherwise—even when the stipulated organization is preserved (Searle 2000; Seth 2025). But “silicon is not biological” is not yet an explanation; the relevant lost property must be identified and defended. The same burden applies in reverse. Functional similarity at an arbitrarily abstract level is insufficient; an organization-sensitive theory must specify which preserved organization is relevant to the mental kind. The Synthetic Pain Continuum therefore places both sides under pressure: neither “different material” nor “same function” settles the attribution by itself.
The disagreement need not itself be gradual: a theory may predict continuous change, a threshold, an effect from the first replacement, or even deny that the stipulated preservation is physically possible. What matters is not a privileged replacement percentage but which changing property the theory treats as experience-relevant.
The case also returns us to the epistemic asymmetry introduced near the beginning of the paper. From within the thought experiment, the subject continues to say that the experience hurts. From outside, observers see preserved specified organization, reports, behavior, learning, attention, and regulation while the physical realization changes. A theory may nevertheless hold that phenomenology has altered or vanished. That possibility cannot be excluded merely because no third-person measure reveals the change; the Third-Person Self Test already showed why absence of a separately observable “painfulness” is not decisive.
But the converse is equally important. Continued reports do not prove phenomenal continuity. A sufficiently sophisticated nonconscious system might in principle generate pain reports. The thought experiment therefore does not eliminate the other-minds problem. It intensifies it by holding much of the accessible evidence fixed while changing the proposed hidden determinant.
What, then, has the intervention achieved?
The intervention converts a vague biology-versus-function disagreement into a comparative demand: which changing or preserved properties warrant the difference or invariance in attribution?
This is a narrower result than substrate independence, but it is also harder to dismiss. The Synthetic Pain Continuum does not prove that an artificial endpoint suffers. It shows why neither biological difference nor functional similarity can carry that conclusion without an accompanying theory of relevance.
Independent Artificial Attribution
The Synthetic Continuum inherits experiential, causal, and autobiographical continuity from its human starting point. An independently artificial system has no such inheritance, making attribution rather than preservation the central problem.
The question should therefore not be whether a machine that avoids damage is conscious, nor whether a sufficiently convincing pain report proves suffering. Those routes would make the argument circular. Instead, consider an artificial system that, independently of the present discussion, satisfies enough credible indicators to count as a serious candidate for experience. The indicators themselves may be theory-dependent and revisable. The important point is that experiential candidacy is established, if at all, by evidence external to the system’s pain-like behavior.
Current AI consciousness can therefore be bracketed. Indicator approaches treat experiential candidacy as theory-dependent rather than behaviorally self-evident, while artificial-pain work decomposes candidate functions rather than reading pain directly from resemblance (Butlin et al. 2026; Henke 2026). Our route differs in direction: the artificial case arrives only after external harm, temporal integration, control, architecture, and substrate have already been varied. Here developmental and material continuity with the starting human finally disappears.
Consider such a system embodied in a robot with an articulated arm.
The arm sustains structural damage. Sensors register impairment. The system represents the affected component as belonging to its own embodied model. The damage interferes with current goals and receives high processing priority. The system attempts to protect or repair the arm. Those attempts fail. Attention and computational resources continue to be redirected toward the unresolved condition. The system learns to avoid circumstances associated with the damage and predicts that continued use will worsen it.
None of these facts should initially be called pain.
A thermostat detects discrepancy. A control system can prioritize an error. A reinforcement learner can avoid states associated with negative reward. A robot can protect damaged hardware. Repetition does not transform those mechanisms into suffering merely by occurring often.
The question becomes more interesting as organization becomes richer.
Suppose the system represents not merely “component X is damaged” but “my arm is damaged.” It incorporates the condition into a self-model. It represents the impairment as obstructing states it is organized to attain. It attempts correction and repeatedly fails. It predicts continued failure. It remembers previous unsuccessful attempts. It changes future behavior to avoid recurrence.
Now reuse the earlier interventions.
Give one version almost no cross-episode integration while another possesses detailed autobiographical memory. Give another strong future simulation. Allow one system to terminate the condition immediately while another has no effective means of doing so. Preserve the same present damage-related signal while varying perceived control, memory, persistence, prediction, and self-relevance.
These variants would almost certainly differ functionally. The system with autobiographical integration can represent “this keeps happening to me.” The predictive system can represent future recurrence. The uncontrollable system can repeatedly encounter failed correction. The self-modeling system can represent the impairment as a condition of itself rather than merely as an external variable.
But functional enrichment still does not answer the central question.
One influential artificial-suffering proposal requires consciousness, a phenomenal self-model, negative valence, and phenomenal transparency, with valence tied to frustrated preferences represented as one’s own (Metzinger 2021). This captures a self-involving dimension that simple damage detection misses. Yet “negative valence” cannot be allowed to end the analysis. If negative valence means simply that the state feels bad, then saying that suffering requires negative valence amounts to saying that suffering requires feeling bad. The statement may be true while explaining nothing.
Computational accounts can explain how frustration, unpredictability, uncontrollability, error signals, and failed correction sustain a high-priority regulatory state (Hyvärinen 2024). Such mechanisms are relevant to the first two explanatory levels developed here: regulation and affective organization.
They do not automatically settle the third: error minimization, conflicting objectives, repeated failure, self-modeling, and memory can enrich the architecture without by themselves demonstrating that any state is bad to undergo.
Even granting computational consciousness would leave a valence-grounding problem: what makes a state genuinely good or bad for the system rather than merely something an objective function ranks, penalizes, or avoids (Wasserziehr 2026)?
A negative reward, penalty, error, or loss does not establish phenomenal significance merely because its mathematical sign is negative.
The artificial case concentrates the earlier problems without the evidential comfort of biological continuity. Protection, intelligence, sophisticated behavior, and error signals do not establish phenomenology. If the system lacks credible indicators of experience, pain-like behavior is insufficient; if it becomes a serious experiential candidate, the question becomes which features of the damage-related state warrant aversive attribution. The same comparative relevance burden now applies under a much weaker evidential background.
Suppose, finally, that every regulatory and organizational feature we have discussed were specified with extraordinary precision. We knew how the system represented damage, how it modeled itself, how it integrated memory, how it predicted future states, how control altered processing, how attention was allocated, how learning changed behavior, and how all of these processes interacted over time.
Would we then know why any of it felt bad?
That question is not a defect in the artificial case. It is the reason the artificial case is not the climax. Once the unfamiliarity of the machine has been stripped away, the same question returns to the human case with which we began.
Before confronting it directly, however, the processes accumulated across the paper need to be put back together. We have spent the argument decomposing aversive experience. The next task is reconstruction.
6. Reconstructing Aversive Experience
Decomposition has a cost. Once a familiar phenomenon has been divided into sensation, appraisal, attention, motivation, memory, prediction, control, self-relevance, learning, and temporal integration, the resulting list can become less explanatory than the ordinary word it replaced. A person does not ordinarily experience ten independent variables. The variables interact within one evolving system.
The task is therefore to reconstruct without simply restoring the original category unchanged.
The pieces need not be independent to be analytically useful. Existing integration, component-process, predictive/interoceptive, self-model, and dynamical approaches all provide ways of treating bodily and affective states as coordinated, history-dependent organization rather than one essential pain component (Scherer 2009; Fulkerson 2026; Seth 2013; Gerrans 2024; Ho and Ryan 2026). The preceding interventions motivate only the weaker claim needed here: control, prediction, memory, self-relevance, and prior states can change the configuration into which a new signal arrives, so trajectories and interactions may matter in addition to isolated variables.
We can now introduce a deliberately nontechnical analogy.
Imagine a chamber with multiple inputs around its boundary. Each input releases a gas with a distinctive color gradient. One input corresponds to sensory or interoceptive information. Others correspond to attention, memory, prediction, perceived control, self-relevance, preference conflict, and motivational urgency.
The labels on the valves are analytical distinctions. They do not imply that nature contains literal independent channels corresponding exactly to our vocabulary.
Once gases enter the chamber, they mix.
A previous mixture changes the effect of the next input. An input can stop while its influence remains. New information arrives into a system already shaped by memory, expectation, mood, attention, and prior attempts at regulation. A representation of the future can alter the chamber before the represented event occurs. A past event can cease to be explicitly remembered while changes it produced remain.
The color visible at any moment therefore belongs to the global configuration, not simply to the last valve opened.
Suppose the resulting mixture is described as brown. It would be a mistake to infer that there must therefore be a separate brown-gas input hidden somewhere around the chamber. Brown may describe the configuration produced by interacting inputs.
The analogy suggests a possibility about suffering. A state categorized as suffering need not correspond to one dedicated “suffering signal.” It may correspond to a characteristic configuration or trajectory involving several interacting processes.
But the analogy must stop there.
It does not establish that phenomenal suffering is nothing over and above the configuration. It does not show that a particular mixture necessarily feels bad. It does not prove functionalism, emergence, constructionism, or any theory of consciousness. The chamber is a way of visualizing organization already motivated independently; it carries no additional evidence of its own.
Abstractly, later states depend not only on new input but on the system’s current organization and ongoing regulation. The point is simply history dependence: the system receiving the next signal has already been changed by what preceded it.
Reconstruction avoids two opposite mistakes. The absence of one indivisible pain mechanism does not make integrated patterns unreal, but interaction should not collapse into the empty claim that “everything affects everything.” A substantive account must still specify which dimensions matter, how they interact, and what directional relations are proposed. The present framework is a scaffold for comparing those commitments, not a completed causal model.
We can now distinguish three explanatory levels that have sometimes been blurred.
At the first level are regulatory dynamics. What discrepancy is detected? Which states does the system attempt to maintain or avoid? How are resources allocated? What actions are taken? What is learned when correction succeeds or fails?
At the second level is affective organization. How do bodily information, appraisal, attention, memory, prediction, control, self-relevance, and action tendencies become coordinated into the evolving state of the subject?
At the third level is phenomenal aversion. Why, if at all, is this configuration bad to undergo?
Much of the preceding paper has increased resolution at the first two levels. Damage can be separated from representation of damage. Immediate aversion can be separated from temporal integration. Control can change the organization without changing the stimulus. Different nervous architectures can support converging evidence of affective significance. Substrate and organization can be varied conceptually. Artificial systems can instantiate increasingly rich regulatory and self-representational structures.
Suppose, then, that the chamber were completely mapped.
Suppose we knew every relevant input, every interaction, every memory effect, every prediction, every control relation, every self-representation, every learning process, and every transition through time. Suppose our description predicted behavior perfectly.
Would that description tell us why this configuration feels terrible?
Different philosophical theories answer differently. Some deny that another ingredient is required. Some identify the phenomenal state with the physical or functional organization. Some locate unpleasantness in evaluative or imperative structure. Some argue that the remaining gap is conceptual. Others insist that phenomenal character remains unexplained.
The reconstruction therefore does not end the inquiry.
It tells us where the inquiry has arrived.
We began with a seemingly indivisible fact: this hurts. We dismantled many assumptions surrounding it and reconstructed a dynamically organized system. If something remains unexplained, we can now ask what that remainder actually is rather than hiding it inside the original word “pain.”
7. The Phenomenal Remainder
Suppose the reconstruction succeeded beyond anything currently possible.
Suppose every relevant regulatory process were known. We could describe the sensory input, bodily regulation, appraisal, attention, memory, prediction, self-model, perceived control, learning, action tendencies, and every interaction among them. We could trace the system through time and predict its behavior. We could explain why it withdraws, why it protects an injured body part, why it cannot stop attending to the state, why previous episodes alter future responses, and why loss of control changes the trajectory.
Would anything remain unexplained when the subject says: this is awful?
The answer depends on what one thinks unpleasantness is.
Evaluative accounts locate unpleasantness partly in representing a disturbance as bad for the subject (Bain 2013; Mitchell 2025).
The pressure is whether evaluative content explains phenomenal unpleasantness or merely redescribes it: if “bad” is defined phenomenally the explanation risks circularity; if defined independently, the account must explain why that content has the relevant affective character.
Imperative accounts instead emphasize directive structure—protect, stop, escape—connecting unpleasantness to the action-guiding organization highlighted by failed correction and loss of control (Klein 2007).
The parallel pressure is that directive structure can exist in sophisticated control without obviously being miserable; the theory must therefore explain why the relevant command structure constitutes unpleasantness rather than merely action guidance.
Motivational accounts similarly connect suffering with minding a state, though desire and unpleasantness can dissociate, leaving open whether motivation constitutes, amplifies, or responds to the aversion.
Functionalist accounts identify the relevant mental state with a richer causal organization across attention, motivation, learning, action, memory, and regulation, naturally accommodating architectural or material variation when that organization is preserved (Chalmers 1995).
The familiar pressure remains whether such organization entails a particular felt character; Chalmers’s replacement arguments make divergence difficult to sustain without empirically demonstrating organizational invariance (1995).
Identity approaches reject the demand for an additional ingredient more directly: phenomenal aversion may simply be identical with a physical or dynamical state, so asking for painfulness after specifying that state may mistakenly demand two things where there is one.
The first-person character is not as transparently derived from third-person description as familiar scientific identities, but the view shows why “Where is the phenomenal property?” need not imply another spatial object.
Phenomenal-concept strategies locate that opacity partly in distinctive concepts of experience: the underlying state could be physical while first- and third-person routes represent it differently (Stoljar 2005; Papineau 2007). The Third-Person Self Test fits this possibility, but the strategy relocates rather than automatically eliminates the burden: it must explain why the conceptual relation generates the apparent gap without reproducing it (Chalmers 2007).
Illusionism instead denies that the apparently intrinsic phenomenal property exists as introspection presents it: the aversive organization and its introspective representation remain real, while the theory owes an explanation of why they generate the powerful appearance of immediately given phenomenal awfulness (Frankish 2016).
Phenomenal realist and anti-reductionist views instead hold that phenomenal character is not exhausted by functional description. They thereby inherit the attribution questions developed earlier: which systems instantiate the relevant property, which transformations preserve it, and what evidence warrants extending it beyond the first-person case?
Evaluative, imperative, motivational, functional, identity, phenomenal-concept, illusionist, and realist approaches relocate the remaining explanatory burden in different ways. The comparison is diagnostic rather than a scorecard: the progressive interventions reveal where each account places that burden.
Further decomposition alone cannot decide among those answers. Each position owes an account of why its chosen feature of the reconstructed state explains, or fails to explain, its being bad to undergo.
8. Conclusion: This Hurts, Revisited
First-person aversion remains immediate even after its conditions have been separated. A subject’s “this hurts” can be accompanied by damage, expectation, memory, loss of control, and fear of persistence. Their coexistence does not make them identical. The interventions identify which differences bear on the experience and which bear on its interpretation, duration, or attribution.
These distinctions can matter practically even while the metaphysics remains unresolved. Attention, expectation, learning, emotion, and perceived control can modulate pain experience without making injury imaginary or implying that pain can simply be thought away (Wiech 2016; Peerdeman et al. 2016; Löffler et al. 2018). Their clinical usefulness is an empirical question beyond this paper. Here they show why immediate aversion, expectation, memory, interpretation, and anticipated persistence should not be collapsed into a single target.
At every transition, the same comparative relevance burden applies. A change in attribution requires an independently motivated experience-relevant difference; preservation of attribution across architectural or physical change requires an independently motivated basis for invariance. Similarity alone does not establish experience, and difference alone does not defeat its attribution.
The method leaves two connected questions: what makes an organized state bad to undergo, and what warrants attributing that property when first-person access is absent. Neither biological familiarity nor artificial unfamiliarity answers them by itself.
REFERENCES
Bain, David. 2013. “What Makes Pains Unpleasant?” Philosophical Studies 166: 69–89. https://doi.org/10.1007/s11098-012-0049-7.
Birch, Jonathan, Alexandra K. Schnell, and Nicola S. Clayton. 2020. “Dimensions of Animal Consciousness.” Trends in Cognitive Sciences 24(10): 789–801. https://doi.org/10.1016/j.tics.2020.07.007.
Butlin, Patrick, Robert Long, Tim Bayne, Yoshua Bengio, Jonathan Birch, David Chalmers, Axel Constant, George Deane, Eric Elmoznino, Stephen M. Fleming, Xu Ji, Ryota Kanai, Colin Klein, Grace Lindsay, Matthias Michel, Liad Mudrik, Megan A. K. Peters, Eric Schwitzgebel, Jonathan Simon, and Rufin VanRullen. 2026. “Identifying Indicators of Consciousness in AI Systems.” Trends in Cognitive Sciences 30(6): 488–501. https://doi.org/10.1016/j.tics.2025.10.011.
Chalmers, David J. 1995. “Absent Qualia, Fading Qualia, Dancing Qualia.” In Conscious Experience, edited by Thomas Metzinger, 309–328. Imprint Academic.
Chalmers, David J. 2007. “Phenomenal Concepts and the Explanatory Gap.” In Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and Physicalism, edited by Torin Alter and Sven Walter, 167–194. Oxford University Press.
Coninx, Sabrina. 2020. “Pain, Amnesia, and Qualitative Memory: Conceptual and Empirical Challenges.” Journal of Consciousness Studies 27(11–12): 126–133.
Coninx, Sabrina. 2022. “A Multidimensional Phenomenal Space for Pain: Structure, Primitiveness, and Utility.” Phenomenology and the Cognitive Sciences 21: 223–243. https://doi.org/10.1007/s11097-021-09727-0.
Coninx, Sabrina. 2023. “The Notorious Neurophilosophy of Pain: A Family Resemblance Approach to Idiosyncrasy and Generalizability.” Mind & Language 38(1): 178–197. https://doi.org/10.1111/mila.12378.
Coninx, Sabrina. 2024. “Pain Philosophy: Recent Debates and Future Challenges.” Philosophy Compass 19(4): e12981.
Corns, Jennifer. 2016. “Pain Eliminativism: Scientific and Traditional.” Synthese 193(9): 2949–2971. https://doi.org/10.1007/s11229-015-0897-8.
Crook, Robyn J. 2021. “Behavioral and Neurophysiological Evidence Suggests Affective Pain Experience in Octopus.” iScience 24(3): 102229. https://doi.org/10.1016/j.isci.2021.102229.
Duval, Alexandre, and Colin Klein. 2026. “Pain Asymbolia Is Probably Still Pain.” Philosophy of Science 93(1): 221–229. https://doi.org/10.1017/psa.2025.10098.
Feinstein, Justin S., Melissa C. Duff, and Daniel Tranel. 2010. “Sustained Experience of Emotion after Loss of Memory in Patients with Amnesia.” Proceedings of the National Academy of Sciences 107(17): 7674–7679. https://doi.org/10.1073/pnas.0914054107.
Frankish, Keith. 2016. “Illusionism as a Theory of Consciousness.” Journal of Consciousness Studies 23(11–12): 11–39.
Fredrickson, Barbara L., and Daniel Kahneman. 1993. “Duration Neglect in Retrospective Evaluations of Affective Episodes.” Journal of Personality and Social Psychology 65(1): 45–55.
Fulkerson, Matthew. 2026. “Pain and Psychological Integration.” Philosophical Psychology 39(2): 605–623. https://doi.org/10.1080/09515089.2024.2425338.
Gerrans, Philip. 2024. “Pain Suffering and the Self. An Active Allostatic Inference Explanation.” Neuroscience of Consciousness 2024(1): niae002. https://doi.org/10.1093/nc/niae002.
Griffith, Aaron M., and Amy Kind. 2024. “Pain Asymbolia Is Not Pain.” Philosophy of Science 91(3): 561–578. https://doi.org/10.1017/psa.2023.167.
Guzmán-Vélez, Edmarie, Justin S. Feinstein, and Daniel Tranel. 2014. “Feelings Without Memory in Alzheimer Disease.” Cognitive and Behavioral Neurology 27(3): 117–129. https://doi.org/10.1097/WNN.0000000000000020.
Henke, Benjamin. 2026. “Studying Artificial Affect: The Case of Pain.” Inquiry 69(6): 2896–2917. https://doi.org/10.1080/0020174X.2025.2556747.
Ho, Tim, and Mark Ryan. 2026. “Chronic Pain as a State-Constrained Brain Network Disorder: A Dynamical Systems Model Integrating Physiological Regulation and Self-Organisation.” Frontiers in Pain Research 7: 1825571. https://doi.org/10.3389/fpain.2026.1825571.
Hyvärinen, Aapo. 2024. Painful Intelligence: What AI Can Tell Us About Human Suffering. 2nd ed. University of Helsinki.
Kahneman, Daniel, Barbara L. Fredrickson, Charles A. Schreiber, and Donald A. Redelmeier. 1993. “When More Pain Is Preferred to Less: Adding a Better End.” Psychological Science 4(6): 401–405.
Klein, Colin. 2007. “An Imperative Theory of Pain.” Journal of Philosophy 104(10): 517–532. https://doi.org/10.5840/jphil2007104104.
Lewis, David. 1980. “Mad Pain and Martian Pain.” In Readings in Philosophy of Psychology, Vol. I, edited by Ned Block, 216–222. Harvard University Press. https://doi.org/10.4159/harvard.9780674594623.c24.
Löffler, Martin, Sandra Kamping, Michael Brunner, Smadar Bustan, Dieter Kleinböhl, Fernand Anton, and Herta Flor. 2018. “Impact of Controllability on Pain and Suffering.” PAIN Reports 3(6): e694. https://doi.org/10.1097/PR9.0000000000000694.
Mendl, Michael, and Elizabeth S. Paul. 2008. “Do Animals Live in the Present? Current Evidence and Implications for Welfare.” Applied Animal Behaviour Science 113(4): 357–382. https://doi.org/10.1016/j.applanim.2008.01.013.
Metzinger, Thomas. 2021. “Artificial Suffering: An Argument for a Global Moratorium on Synthetic Phenomenology.” Journal of Artificial Intelligence and Consciousness 8(1): 43–66. https://doi.org/10.1142/S270507852150003X.
Mitchell, Jonathan. 2025. “Towards Affective-Evaluativism: The Intentional Structure of Unpleasant Pain Experience.” The Philosophical Quarterly 75(2): 693–717. https://doi.org/10.1093/pq/pqae002.
Mogensen, Andreas L. 2026. “Welfare and Felt Duration.” Noûs 60(3): 563–575. https://doi.org/10.1111/nous.70022.
Montero, Barbara Gail. 2020. “What Experience Doesn’t Teach: Pain Amnesia and a New Paradigm for Memory Research.” Journal of Consciousness Studies 27(11–12): 102–125.
Nielsen, Tore A., Diane L. McGregor, Antonio Zadra, Diann Ilnicki, and Lucie Ouellet. 1993. “Pain in Dreams.” Sleep 16(5): 490–498. https://doi.org/10.1093/sleep/16.5.490.
Papineau, David. 2007. “Phenomenal and Perceptual Concepts.” In Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and Physicalism, edited by Torin Alter and Sven Walter, 111–144. Oxford University Press.
Peerdeman, Kaya J., Antoinette I. M. van Laarhoven, Madelon L. Peters, and Andrea W. M. Evers. 2016. “An Integrative Review of the Influence of Expectancies on Pain.” Frontiers in Psychology 7: 1270. https://doi.org/10.3389/fpsyg.2016.01270.
Polger, Thomas W., and Lawrence A. Shapiro. 2016. The Multiple Realization Book. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198732891.001.0001.
Rainville, Pierre, Benoît Carrier, Robert K. Hofbauer, M. Catherine Bushnell, and Gary H. Duncan. 1999. “Dissociation of Sensory and Affective Dimensions of Pain Using Hypnotic Modulation.” Pain 82(2): 159–171. https://doi.org/10.1016/S0304-3959(99)00048-2.
Raja, Srinivasa N., Daniel B. Carr, Milton Cohen, Nanna B. Finnerup, Herta Flor, Stephen Gibson, Francis J. Keefe, Jeffrey S. Mogil, Matthias Ringkamp, Kathleen A. Sluka, Xue-Jun Song, Bonnie Stevens, Mark D. Sullivan, Perri R. Tutelman, Takahiro Ushida, and Kyle Vader. 2020. “The Revised International Association for the Study of Pain Definition of Pain: Concepts, Challenges, and Compromises.” Pain 161(9): 1976–1982. https://doi.org/10.1097/j.pain.0000000000001939.
Rebello, Travis. 2026. “Well-Being and Psychological Continuity.” Philosophical Studies 183: 2243–2263. https://doi.org/10.1007/s11098-026-02537-7.
Scherer, Klaus R. 2009. “The Dynamic Architecture of Emotion: Evidence for the Component Process Model.” Cognition and Emotion 23(7): 1307–1351. https://doi.org/10.1080/02699930902928969.
Schnell, Alexandra K., Heather Browning, Andrew Crump, Charlotte C. Burn, and Jonathan Birch. 2026. “Sentience in Cephalopod Molluscs: An Updated Assessment.” Biological Reviews 101(3): 1311–1333. https://doi.org/10.1002/brv.70125.
Searle, John R. 2000. “Consciousness.” Annual Review of Neuroscience 23: 557–578. https://doi.org/10.1146/annurev.neuro.23.1.557.
Seth, Anil K. 2013. “Interoceptive Inference, Emotion, and the Embodied Self.” Trends in Cognitive Sciences 17(11): 565–573. https://doi.org/10.1016/j.tics.2013.09.007.
Seth, Anil K. 2025. “Conscious Artificial Intelligence and Biological Naturalism.” Behavioral and Brain Sciences 49: e315. https://doi.org/10.1017/S0140525X25000032.
Stoljar, Daniel. 2005. “Physicalism and Phenomenal Concepts.” Mind & Language 20(5): 469–494. https://doi.org/10.1111/j.0268-1064.2005.00296.x.
Talbot, Kerwin A., Victoria J. Madden, Sara L. Jones, and Lorimer Moseley. 2019. “The Sensory and Affective Components of Pain: Are They Differentially Modifiable Dimensions or Inseparable Aspects of a Unitary Experience? A Systematic Review.” British Journal of Anaesthesia 123(2): e263–e272. https://doi.org/10.1016/j.bja.2019.03.033.
Wasserziehr, Jan Henrik. 2026. “Are Conscious Machines Valuers?” AI & Society 41: 4957–4969. https://doi.org/10.1007/s00146-026-02959-1.
Wiech, Katja. 2016. “Deconstructing the Sensation of Pain: The Influence of Cognitive Processes on Pain Perception.” Science 354(6312): 584–587. https://doi.org/10.1126/science.aaf8934.
Zadra, Antonio L., Tore A. Nielsen, Anne Germain, Gilles Lavigne, and Don C. Donderi. 1998. “The Nature and Prevalence of Pain in Dreams.” Pain Research and Management 3(3): 155–161. https://doi.org/10.1155/1998/946171.