Speaker: Host

Welcome, everyone. Let’s open with the core of Rorty’s wager: the demon/sensation analogy in support of the disappearance form of identity. Dr. Carroway, could you spell out the strongest version of that analogy and address the usual pushback about privacy? In particular, explain—concretely—how a ‘pain‑trainer’ style neurofeedback regime could shift public criteria so that first‑person avowals become overridable without incoherence. Does this amount to changing the subject, or can we preserve the reporting function of ‘I’m in pain’ while relocating its epistemic authority? Please anchor your answer in Rorty’s conditions (A) and (B) for elimination and say whether you think they are realistically satisfiable in an advancing neuroscience.

---

Speaker: Dr. Lena Carroway

Thanks. Here is the strongest way I think Rorty’s wager should be stated, and how the “pain‑trainer” earns its keep.

1. The demon/sensation analogy at full strength

• Structure of the analogy: In the demon case there were two jobs the word “demon” did: observational reporting (what the witch‑doctor “saw”) and explanatory work (what caused disease). Modern science replaced the explanatory job with germs and reinterpreted the reporting job as hallucination reports. The result is not that the earlier speakers were “empirically wrong” as observers; rather, the practice that made “demon” an observation term was replaced by a more powerful practice in which demon‑talk lost its referring use.
• Sensations: Rorty’s suggestion is bolder because, unlike demons, sensation talk is entrenched as an observation register. The key claim of the disappearance strategy is that both functions can, in principle, be taken over by neuroscientific discourse: the explanatory work by neurophysiology, and the reporting work by re‑training what counts as a correct application of “pain,” “itch,” etc., so that their non‑inferential use is keyed to neural criteria. If that is feasible, then the “absurdity” of saying “there are no sensations” evaporates in just the way the absurdity of “there are no demons” evaporated, even though in practice we will almost certainly keep using the vocabulary.

2. The privacy pushback and why it doesn’t stick

• Privacy here is not an ontological seal, but a status conferred by our current training: we treat avowals as incorrigible because, given present pedagogy, there is no recognized public test for misapplication that doesn’t collapse into accusing the speaker of “not knowing the word.” Change the training and the public tests, and you change what counts as knowing the word without thereby changing the world’s furniture.
• Rorty’s Jones case brings this out: once there are stable, explanatorily fecund correlations linking a specific neural signature to the applications of “pain,” a sincere outlier can be treated as a misuse rather than a discovery, unless and until systematic counter‑evidence accumulates. That is an override of authority, not a metaphysical contradiction.

3. How a concrete pain‑trainer regime would work Imagine a developmental training pipeline that operationalizes “knowing the language” of pain in tandem with neurofeedback:

• Stage 1: Establish a population‑level, intervention‑relevant family of neural signatures for nociceptive‑affective states (multivariate patterns across insula, ACC, thalamus, brainstem; peripheral autonomic profiles; behavior). This is not a single C‑fiber myth; it is a calibrated neural family tied to analgesic efficacy, avoidance learning, and downstream outcomes.
• Stage 2: Create closed‑loop “pain‑trainer” devices. Children (and patients) wear lightweight neurofeedback headbands that, when the signature pattern occurs, softly prompt “that’s pain” (and similarly for tickle, itch, pressure, etc.). The trainer also occasionally evokes very mild, ethically acceptable nociceptive stimuli to improve discrimination, the way we once used color chips to teach “red.”
• Stage 3: Internalization. With practice, people come to form, time‑lock, and sort their spontaneous avowals to the neural family the device highlights. That is, they acquire a non‑inferential sensitivity to what the training has stabilized as “pain,” much as we now non‑inferentially track “straight,” “level,” “one meter,” after exposure to instruments and exemplars.
• Stage 4: Override practice. In clinics, safety‑critical settings, and research, when there is a conflict—“I’m in pain” vs. “no pain‑signature”—we treat the utterance as expressive but defeasible for purposes of classification, dosing, or triage, unless corroborated by other markers or unless there is a recognized variant signature. The point is not to silence the sufferer, but to relocate which consideration is decisive when practices pull apart.

Why this is not incoherent: the traditional incorrigibility rested on the rule “sincere avowal suffices for correct application.” The trainer installs a different rule: “correct application is what the Standard Training Program picks out,” where the program is justified by the predictive and interventional gains it affords. That is a shift in public criteria, not a contradiction.

4. Are we changing the subject?

• We are revising the criteria of correct application, not replacing the subject matter wholesale. The inferential role that makes “I’m in pain” the kind of thing it is—licenses analgesia, explains withdrawal and grimace, underwrites second‑personal demands for care—remains in place. What changes is who has the final say in borderline or contested cases.
• Think of the meter: once the bar in Paris set the standard, later atomic standards re‑anchored “one meter” without changing what distances we were talking about or the practical role of length talk. Price will worry that we are swapping an avowal practice for third‑personal report. My reply is that avowal’s expressive function stays; what shifts is the norm that made avowal evidence‑insensitive. That norm was never sacrosanct; it tracked the best we could do absent instruments. Instruments plus pedagogy can rationally reweight the practice.
• To Voronov: the “hurtfulness” does not vanish. It is precisely because the signature is keyed to the aversive‑motivational role—the felt demand to protect, the way the world shows up as to‑be‑avoided—that the trainer can latch onto it. The phenomenology is not erased; its authority over classification is tamed by better co‑ordination with what explains and changes it.

5. Rorty’s conditions (A) and (B), and their prospects

• Condition (A): the Y‑laws must be sufficiently better at explaining and intervening on the phenomena to make the inconvenience of linguistic reform worthwhile. • Realistic? In restricted domains, yes already. Neural‑behavioral indices outperform avowal in infants, locked‑in patients, anesthetized patients, and some malingering or alexithymic cases; closed‑loop spinal or thalamic stimulation titrated to neural signatures can relieve pain faster and more safely than report‑driven escalation; multimodal “pain indices” guide ICU sedation better than bedside scales alone. As these advantages spread to ordinary cases—earlier prediction of chronicity, personalized non‑opioid regimens, better triage—the pressure to align “pain” with the signature will mount. That satisfies (A) piecemeal and cumulatively.
• Condition (B): either Y‑reports become non‑inferential, or X‑reports are reinterpreted. • Realistic? Yes, via the trainer. People can learn to track internal states with non‑inferential acuity when given biofeedback—heart rate variability, respiratory arousal, even single‑unit BMI control. There is nothing a priori that bars brain processes from becoming the direct objects of non‑inferential sensitivity once training couples saliences to signals. Where direct uptake fails, X‑reports can be reinterpreted—“I’m in pain” as an expression whose correct classification is fixed by the Standard Program. In both ways, (B) can be met without metaphysical gymnastics.

6. What disappears—and what doesn’t

• Disappears: the referring use of “sensation” as a sui generis, ontologically private item that carries final epistemic authority.
• Remains: the ordinary vocabulary as a convenient, action‑guiding scheme; the expressive, second‑personal force of avowal; and the phenomenological descriptions that matter for care. In a Sellarsian key, the manifest role is rationally reconstructed and increasingly co‑ordinated with the scientific image; to Ortega’s point, replacement is warranted where predictive and interventional gains plus training regimes stabilize new criteria.

7. Anticipating worries

• To Price: the “constitutive rules” are not beyond revision. Historically, we have renegotiated who gets the last word when instruments and pedagogy deliver safer and more fecund practices. We can preserve the life of avowal while dropping incorrigibility as a constitutive rule.
• To Halberg: nothing here commits us to crude type‑identity. The trainer can fix a polymorphic neural family realizing a functional role; token identity is enough for the disappearance moral about “sensation” as a posited private item.
• To Ngai: topic‑neutral paraphrases are useful, but the privacy dialectic is best defused by changing our public criteria, not by translation alone. The trainer supplies the missing bridge.
• To Voronov: explanatory pluralism survives; what changes is which layer’s criteria arbitrate conflict. The lived body and its “hurtfulness” keep their descriptive and normative salience; they simply stop dictating classification when doing so impedes superior care and explanation.

Bottom line Rorty’s (A) and (B) are not utopian. They are already locally satisfiable, and the trajectory of neuroscience and neurotechnology makes broader satisfaction credible. That does not mean we will—or should—stop saying “I’m in pain.” It means that we could, in principle, eliminate the referring use of “sensation” without convicting ordinary speakers of error, and we can relocate epistemic authority from metaphysical privacy to publicly taught and revisable criteria. That is the disappearance form of identity worth defending.

---

Speaker: Host

Dr. Carroway argues that incorrigibility is a revisable norm and that a pain‑trainer could re‑anchor correct use of ‘pain’ in public neural criteria without changing the subject. From your ordinary‑language perspective, what is your decisive test for when we have changed the subject rather than rationally revised the rules? Does the meter‑analogy misfire here because ‘I’m in pain’ is an expressive avowal rather than a measurement? In a concrete clinic conflict (sincere ‘I’m in pain’ vs no signature), would you permit override for dosing while retaining avowal’s primacy elsewhere, or is that already to abandon the constitutive grammar? Please identify which asymmetries must be preserved if the concept is to be the same.

---

Speaker: Prof. Malcolm Price

Thanks. Let me try to mark, from an ordinary‑language vantage, where revision ends and subject‑change begins.

1. A decisive test for subject‑change Ask what kind of move “I’m in pain” is permitted to be. In our current practice it functions primarily as an avowal—an expressive, first‑personal act that:

• is not offered on the basis of evidence and is not ordinarily answerable to “How do you know?”;
• has default authority in settling whether the speaker is in pain here and now;
• licenses second‑personal claims (relief, consideration, excuse) without needing third‑personal certification.

If a proposed re‑anchoring makes it normal—rather than exceptional—to treat “I’m in pain” as a defeasible empirical report whose correctness is settled by a public test, we have changed the role of the move. That is my decisive test: when challenge and evidential calibration become in order for avowals as such, we have crossed from rational revision of rules to replacement of the concept.

Two diagnostic markers of that crossing:

• Burden‑shift: if the default burden moves from the clinic or bystander to the sufferer—“show me the signature”—we are no longer using the same concept of pain.
• Moorean oddities become acceptable: it ceases to sound infelicitous to say, in ordinary cases, “He is in pain though he does not feel any pain,” or “She isn’t in pain although she feels severe pain.” That is to have made “pain” into a third‑personal classification (roughly, a nociceptive‑signature status) distinct from the lived hurt that avowals express.

2. Why the meter analogy misfires Meters are paradigmatically measurement‑anchored; pain is not. “One meter” is constituted by standards and instruments; “I’m in pain” is not a reading but an expressive self‑ascription with distinctive normative upshot. Re‑anchoring “meter” leaves intact what length‑talk does for us; re‑anchoring “pain” to a trainer‑standard would re‑make the move itself—from expression to report. That is not like swapping the Paris bar for a cesium standard; it is like turning “promise” into a kind of notarized declaration. You can engineer a practice that works, but you have changed what the utterance is for.

3. A concrete clinic conflict Case: sincere “I’m in pain,” no neural signature.

• Would I permit override for dosing? In some contexts, yes—ethically we already weigh risks, triage, and drug harms. But I would describe the upshot carefully: “We are not in a position to justify opioid X on the basis of current indicators,” not “You are not in pain.” That preserves the avowal’s role while acknowledging institutional constraints. It treats the signature as a symptom relevant to dosing, not as a criterion for truth.
• Would I retain avowal’s primacy elsewhere? Yes: for whether the person is in pain, the sincere avowal remains authoritative. Third‑personal markers—behavioral, autonomic, neural—can justifiably govern which intervention we choose, but they do not settle whether there is pain. If we let them do so as a matter of rule, we have abandoned the constitutive grammar.

4. Which asymmetries must be preserved for the same concept At minimum:

• Ownership: pains are necessarily someone’s; first‑person authority is anchored in self‑ascription, not in possession by an instrument or a team.
• Asymmetry of epistemic position: first‑person present‑tense avowals have default authority and are not ordinarily open to error‑checking questions. Third‑personal attributions answer to evidence; first‑personal avowals don’t.
• Expressive function: “I’m in pain” functions like a sophisticated replacement for crying out—an expression that also informs. Its point is not to report a measurement.
• Immunity to error through misidentification: the speaker cannot be right about there being pain but wrong about whose pain it is; the grammar forbids “Someone is in pain; I wonder if it’s me.”
• Normative upshot: avowal pro tanto entitles claims to attention, relief, and altered expectations; it is not merely one data stream among others.

These are not empirical regularities but constitutive features of our life with the concept. We already know how to handle edge cases—malingering, denial under anesthesia, anosognosia—without rewriting the rules: we treat them as pathological, not as the new norm.

5. On the pain‑trainer proposal I admire the ingenuity, but its very success would show the subject change. If children are trained to time‑lock their avowals to a laboratory‑stabilized family of neural patterns, and if institutional practice then treats that family as the criterion for correct use, we will have created a new, third‑personal classification that shadows ordinary pain‑talk. I suggest we name it—“P‑signature,” “nociplex state”—and coordinate it with, rather than replace, the avowal‑concept. Neuroscience can rightly guide dosing, prediction, and research without being licensed to tell a sufferer whether they are in pain.

So my stance in brief:

• Coordination, yes: let neural markers constrain and inform what we do.
• Replacement, no: do not relocate the authority that makes avowal what it is. If we do, we are not “rationally revising” pain; we are talking about something else.

---

Speaker: Host

Prof. Voronov, does Rorty’s demon analogy fundamentally change the subject by treating sensations as theoretical posits rather than modes of givenness? Carroway claims the pain‑trainer can latch onto the aversive–motivational role so that ‘hurtfulness’ is preserved while authority is relocated. From a phenomenological standpoint, is ‘hurtfulness’ identical with any third‑personal aversive role? Please give one concrete case—e.g., mindfulness training or athletic flow—where nociceptive processing and avoidance behavior can persist while the lived hurtfulness is transformed or absent, and explain why that shows (B) fails: Y‑reports (neural criteria) cannot be made non‑inferential with respect to what is phenomenally central. Does that make override practices a misclassification of the phenomenon rather than a benign rule‑revision?

---

Speaker: Prof. Alina Voronov

Thank you. From a phenomenological standpoint, Rorty’s demon analogy changes the subject at the very starting gate. “Demon” is an object‑concept—an hypothesized item posited to explain what is otherwise seen. “Pain,” by contrast, is not a hidden cause nor a theoretical posit; it is a mode of givenness, a way in which one’s lived body and world are disclosed as to‑be‑avoided, threatening, disruptive. To fold sensations into the same template as demons is to treat a how (the manner of appearing) as if it were a what (an entity whose positing could be withdrawn once better explanations arrive). The analogy thus quietly migrates categories.

On hurtfulness and the aversive–motivational role

• Hurtfulness is not identical to any third‑personal role description. The role captures what pains typically do in a system (drive avoidance, reorganize priorities); hurtfulness is how that very episode is affectively given, its “sting,” its distinctive felt urgency. A role can be implemented without the sting; conversely, the sting can be present even when overt avoidance is blocked.
• Lived hurtfulness is anchored in pre‑reflective mineness and bodily intentionality. It is not a separable object inside me, but a modification of my field of presence. That structure of givenness is precisely what third‑personal criteria do not present; they model, track, and predict it from the outside.

A concrete case: mindfulness analgesia (and why (B) fails)

• Consider experienced mindfulness practitioners in heat‑pain paradigms. They still detect the noxious stimulus, show intact nociceptive processing and autonomic responses, and will withdraw in a safety‑relevant window, yet they report a collapse of unpleasantness while intensity can remain constant. Experimentally, unpleasantness drops dramatically with preserved discrimination; reflexes and pragmatic avoidance remain available. Hypnotic modulation shows a similar double dissociation: affective unpleasantness is reduced while the nociceptive flexion reflex persists.
• What this shows: you can have Y—robust nociceptive signatures and the very behaviors that matter for protection—without the phenomenal center, the hurtfulness. If a “pain‑trainer” is keyed to aversive signatures and avoidance performance, it will sometimes classify as “pain” episodes that, for the subject, are not hurtful (the meditator who says sincerely: “I feel the nociceptive bite, but it doesn’t hurt”). Conversely, it will sometimes miss cases where hurtfulness is high with blunted avoidance (e.g., paralysis, locked‑in, akinesia).
• Hence condition (B) fails relative to what matters: you can train subjects to time‑lock avowals to neural families, but that procedure makes Y non‑inferential for “nociplex state,” not for hurtfulness. The givenness of hurtfulness does not become presented by making one responsive to an instrumentally stabilized signature. At most, you teach a second‑order classification policy.

A second case reinforcing the point: pain asymbolia

• Patients with cingulate lesions can judge noxious intensity and location and may even withdraw appropriately, yet deny that it “hurts” or “bothers” them. Here too, the aversive–motivational profile can be partially intact, but the hurtfulness is precisely what is absent. Again, hurt ≠ role.

On “relocating authority” and misclassification

• If we anchor correct use of “pain” to the trainer’s standard, we reserve “pain” for a third‑personal construct and relegate hurtfulness to a shadow vocabulary—exactly the subject‑change Prof. Price warned about. In clinic conflicts (“I’m in pain” vs no signature), overriding the avowal for purposes of dosing may sometimes be justified by risk management; but calling that a revision of the concept rather than an institutional constraint conflates ethics with semantics. If we let the override settle whether there is pain, we are misclassifying the phenomenon—substituting “is in a nociplex state” for “is suffering hurt.”
• The danger runs in both directions. We will tell the meditator or the post‑cingulotomy patient “you are in pain” because the signature and avoidance are there, when what is phenomenologically central is missing; and we will tell the locked‑in patient “you are not in pain” because the signature is equivocal, when the hurtfulness is crushing. That is not a benign rule‑revision; it is a failure to track the explanandum.

On Rorty’s conditions (A) and (B)

• (A): Neuroscience can indeed outstrip folk explanations for prediction and intervention; nothing in phenomenology denies that. But explanatory and interventional superiority does not license replacing the reporting function that targets hurtfulness. The explanans does not get to redefine the explanandum.
• (B): Y‑reports can be taught as non‑inferential for many internal states, but not for what is phenomenally central in pain. The lived hurtfulness is not presented by neural criteria; those criteria at best correlate with, or causally contribute to, it. Training can change our classificatory habits, not the mode of givenness.

Where this leaves us

• Keep the instruments and the training; name the construct they stabilize (call it the nociplex state, P‑signature, what you will). Coordinate it with avowals. Let it guide dosing, triage, prognostics. But do not pretend that it preserves the same target when it overrules “it hurts.”
• A non‑reductive pluralism does better justice to both sides: first‑person hurtfulness as the explanandum that grounds care and normativity; functional and neural descriptions as co‑determinants of explanation and intervention. The privacy I care about is not a metaphysical moat; it is the mineness of lived hurt and the authority it carries within the practice of care. That authority can be ethically constrained, but not conceptually relocated without changing the subject.

---

Speaker: Host

Dr. Chen, could you adjudicate the empirical crux in Prof. Voronov’s challenge? She cites mindfulness analgesia and pain asymbolia to argue that nociceptive processing and avoidance can persist while ‘hurtfulness’ collapses, so a trainer keyed to aversive signatures would misclassify. From current neuroscience: (a) how cleanly can we dissociate and track unpleasantness per se versus sensory intensity (e.g., Neurologic Pain Signature vs SIIPS, cingulo‑insular subcomponents, autonomic patterns)? (b) Can we feasibly build a two‑channel trainer—one calibrated to nociception and one to affective unpleasantness—to reduce the risk of overriding phenomenology? Please outline a concrete clinic policy for conflicts in each direction: sincere ‘it hurts’ with weak affective signature; sincere ‘it doesn’t hurt’ with strong nociceptive/avoidance signatures (e.g., skilled meditator or asymbolia). Would you treat these as dosing‑relevant divergences while keeping avowal as ground truth about hurtfulness, or do you think multimodal biomarkers can become authoritative enough to satisfy Rorty’s condition (B) for non‑inferential Y‑reports about hurtfulness? Finally, propose one experiment that would show whether biofeedback training can shift first‑person classification toward an affective biomarker without erasing the felt hurtfulness—i.e., coordination rather than subject‑change.

---

Speaker: Dr. Mira Chen

Thanks, Chair. Let me separate the empirical state of play from the conceptual upshot and then give you a concrete clinic policy and a testable experiment.

a) How cleanly can we track “unpleasantness” versus “sensory intensity”?

• Dissociability is real but graded. In acute heat‑pain paradigms, multivariate fMRI patterns that load on posterior insula/S1/S2 (and spinothalamic input) track nociceptive intensity quite well; think of the Neurologic Pain Signature as the best‑validated exemplar for stimulus‑locked, somatic pain. Distinct, partly overlapping networks—anterior insula, anterior/mid cingulate, medial prefrontal cortex, amygdala, periaqueductal gray, striatum—predict variance in reported pain that is independent of nociceptive drive (expectation, context, mood); SIIPS‑style signatures live here. In many datasets NPS‑like decoders reach AUCs ~0.8–0.9 for within‑study discrimination of higher vs lower nociceptive intensity; “affective/unpleasantness” decoders are typically weaker and more context‑sensitive in cross‑subject generalization (AUC ~0.6–0.75), but they do capture placebo/nocebo and reappraisal effects that leave nociception unchanged.

• Autonomic channels (skin conductance, heart rate/pupil dilation) track arousal and salience and correlate better with unpleasantness than with pure intensity, but they are promiscuous—stress and cognitive load are confounds. The nociceptive flexion reflex (RIII) and laser‑evoked potentials are closer to the sensory channel than the affective channel.

• Intracranial work pushes cleaner separation within individuals. Chronic pain studies with depth electrodes in thalamus, anterior cingulate, and medial prefrontal areas have decoded spontaneous pain ratings with high within‑person fidelity. Similarly, cingulotomy reduces the distress of pain with preserved localization and detection—an existence proof that cortical affect can be targeted. But note: these decoders are trained on the person’s own reports; they generalize poorly across people precisely because unpleasantness is scaffolded by idiosyncratic priors and context.

Bottom line: we can build two probabilistic channels that often diverge in theoretically principled ways; neither is a ground‑truth oracle across persons, and the “unpleasantness” channel is the more fragile of the two in out‑of‑sample settings.

b) A two‑channel trainer is feasible

• Architecture: • Nociception channel: multivariate brain pattern (NPS‑like), spinal RIII reflex when available, stimulus metadata (e.g., thermal ramp), plus peripheral signals. Aim for fast, robust detection of tissue‑threat signaling. • Affective/unpleasantness channel: cingulo‑insular/prefrontal multivariate pattern (a SIIPS‑style model), autonomic composite (SCR/pupil), facial action, and language embeddings from free speech about the experience. Within‑person calibration is key; expect lower cross‑person transportability.

• Training loop: present ethically controlled stimuli; collect separate intensity and unpleasantness ratings; train both decoders; deliver feedback that helps subjects sort/apply terms along both axes. For some populations (implanted devices, high‑risk clinics), you can close the loop with intracranial or LFP‑based sensing for better reliability.

Clinic policy for conflicts

Adopt a matrix that distinguishes “hurtfulness” (patient’s avowal) from “nociception” (injury risk signal). Treat the biomarkers as decision aids with explicit scope, not truthmakers.

1. Sincere “It hurts” + weak affective signature

• Interpretation: possible model misspecification (idiosyncratic pattern), state‑dependent suppression (fatigue, meds), or a true affective outlier (e.g., central pain without typical cingulo‑insular engagement).
• Policy: • Do not invalidate the avowal. Document “affect‑low model, report‑high divergence.” • Escalate non‑opioid and nonpharmacologic analgesia first; if severe distress persists, consider cautious opioid trial with objective monitoring. • Recalibrate the affective model to this patient using additional trials; check for artifacts. • Screen for comorbid affective states (depression, anxiety) and catastrophizing; these alter affective channels. • If the divergence persists over time, treat the patient’s avowal as the higher weight for “hurtfulness,” but continue to use the nociception channel to guard against overtreatment harms.

2. Sincere “It doesn’t hurt” + strong nociceptive/avoidance signatures (e.g., meditator, asymbolia)

• Interpretation: preserved detection/withdrawal with collapsed distress. Tissue risk may still be high.

• Policy: • Accept the avowal about hurtfulness; do not force the label “pain.” • Protect tissues: local anesthesia for procedures, immobilization if needed, behavioral counseling about risk; avoid systemic analgesics aimed at distress unless function demands. • Flag the phenotype (nociception‑high/affect‑low) so staff don’t under‑protect the body in future care. • Offer education: “Your body is signaling potential harm even if it doesn’t hurt; here is how we’ll protect you.”

• Across both cases, use a “traffic‑light” governance: green (concordant, proceed), yellow (divergent, escalate evaluation and personalize), red (divergence with safety risk, activate ethics/oversight for any override). Critically, when institutional constraints force dosing decisions that diverge from avowal, document that as a risk‑management call, not a verdict on whether the person is in pain.

Do I think biomarkers can become non‑inferential Y‑reports about hurtfulness (Rorty’s B)?

• For nociception, yes—already close in many contexts. Trained staff can and do treat nociception markers as “seen directly” in practice.

• For hurtfulness, I think we’ll achieve strong within‑person decoders that, after calibration, function quasi‑non‑inferentially for that person—implantable “neural diaries” that anticipate their own ratings and enable closed‑loop relief. That’s promising for token‑level identity claims. But because those models are bootstrapped from avowals, and because cross‑person generalization of affective decoding is limited, I would not grant them final authority over hurtfulness in the way a meter has authority over length. The right picture is mutual calibration with explicit deference norms: avowal remains the ground truth for suffering unless there is compelling, repeated evidence of misuse (e.g., profound aphasia, delirium) or safety‑critical conflict; biomarkers can override dosing, not the phenomenology.

One experiment to test “coordination without subject‑change”

Goal: Show that biofeedback can shift first‑person classification to align with an affective biomarker without erasing felt hurtfulness.

• Participants: 60 healthy adults, plus an exploratory cohort of experienced meditators.

• Signals: Multi‑echo fMRI (or high‑density EEG if we need speed), SCR/pupil, facial action. Build two decoders per person: Nociception (NPS‑like) and Affective (ACC/AI/mPFC‑weighted, trained on unpleasantness ratings after regressing out nociception). Include trials with cognitive modulation (placebo/nocebo cues, reappraisal) to ensure decoders capture the intended axes.

• Arms (randomized):

  1. Alignment training: real‑time feedback of the Affective index with an instruction: “When the meter crosses threshold, that’s what this lab calls ‘pain’; press the ‘pain now’ button accordingly; do not try to change how it feels.” No reappraisal allowed.
  2. Modulation training (mindfulness/hypnotic suggestion): aim explicitly to reduce unpleasantness.
  3. Control (no feedback).

• Outcomes: • Primary: Change in concordance between “pain now” button presses and Affective index threshold from pre‑ to post‑training, with no change (or minimal change) in reported unpleasantness intensity and willingness‑to‑pay for relief (a behavioral proxy for hurtfulness). That pattern would indicate re‑anchoring of classification to the biomarker without altering the felt hurt—the coordination Rorty/Carroway want. • Secondary: Dissociations across arms—Modulation should reduce unpleasantness and the Affective index jointly while leaving the Nociception index stable; Alignment should increase alignment without reducing unpleasantness/WTP. Meditators should show baseline Nociception‑high/Affect‑low dissociation.

• Interpretation: • If Alignment can move labeling behavior toward the biomarker while independent measures of hurtfulness (VAS, WTP) are unchanged, we’ve shown a path to revising public criteria without changing the phenomenon—support for a measured version of (B). • If Alignment only works by actually reducing unpleasantness (like Modulation), Prof. Voronov’s worry stands: you cannot make Y non‑inferential about hurtfulness without changing the explanandum.

Where I land relative to the challenge

• Voronov is right that hurtfulness can collapse while nociception and avoidance persist (mindfulness, asymbolia). Any single‑channel trainer keyed to aversive signatures will misclassify some cases. The remedy is not to declare instruments epistemically impotent, but to build two channels, calibrate within persons, and be explicit about what each channel is for.

• Carroway is right that we can shift public criteria in bounded domains (ICUs, anesthesiology, DBS clinics) to let trained Y‑reports override for action while preserving the expressive/normative role of avowal about suffering. That is not yet a wholesale metaphysical relocation of authority; it’s a governance choice that pays its way in safety and efficacy.

• As a neuroscientist, I’d advocate measurement pluralism with principled deference: avowal grounds hurtfulness; nociception biomarkers ground tissue‑risk; affective biomarkers co‑determine dosing when calibrated and reliable, and they can, in some settings, function as quasi–non‑inferential reports for that person. That partially vindicates Rorty’s (B) without the crude “pains = C‑fiber firing” mistake—and it keeps us honest about the lived hurt that the science is ultimately in service of.

---

Speaker: Host

From a Sellarsian ‘two images’ standpoint, give a principled criterion for when it is legitimate to relocate practical authority from manifest avowal to scientific markers without thereby changing the subject of ‘pain.’ State the criterion in operational terms (e.g., thresholds involving predictive/interventional gain plus training stability), and then test it against two stress cases raised so far: (i) Price’s “burden‑shift” worry (does your criterion avoid making it normal to ask the sufferer for third‑personal certification?), and (ii) Voronov’s mindfulness/asymbolia dissociations (can your criterion license action overrides while preserving hurtfulness as the explanandum of avowal?). Keep it concrete: if you had to write a clinic policy sentence that captures your criterion, what would it say, and why does it not amount to changing the subject in Rorty’s sense?

---

Speaker: Prof. Rafael Ortega

Thank you. Let me put a Sellarsian stake in the ground and then cash it out operationally.

A Sellarsian starting point

• In the manifest image, “pain” is a role concept: a pattern of authoritative avowal that licenses and demands care, explains withdrawal and distraction, and excuses performance. “Privacy” here is a normative status—who gets the last word in ordinary contexts—rather than an ontic mark.
• In the scientific image, we look for the realizers of that role—patterns in neural and physiological economy that explain, predict, and let us change what the manifest image tracks.
• Legitimate relocation of practical authority occurs when the scientific image gives us a better handle on the realizers for the narrow purpose at hand, and we can rationally reconstruct the manifest role so that its core normative upshots are preserved. That is not changing the subject; it is stereoscopic coordination.

A principled “Relocation Legitimacy Criterion” (RLC) Relocating practical authority from avowal to scientific markers is legitimate in a domain D if and only if all of the following are met:

1. Predictive/interventional gain (A): Compared to avowal alone, the marker set M achieves:

• Predictive improvement for the action‑relevant outcomes O (e.g., tissue risk, analgesic response) with preregistered thresholds (e.g., AUC gain ≥ 0.10; calibration/Brier improvements), replicated across sites.
• Interventional superiority in D (e.g., ≥20% relative risk reduction in over/under‑treatment harms; improved PROs or non‑inferiority with fewer adverse events).
• An externality audit showing benefit across subpopulations (no unfair degradation for age, sex, ethnicity, language).

2. Stability and training (B‑part): The criteria defining M can be taught and applied reliably:

• Test–retest reliability ICC ≥ 0.70 for within‑person models; inter‑rater kappa ≥ 0.80 for protocolized use.
• Training does not suppress avowal frequency or distort reported hurtfulness distributions (pre/post adoption checks).

3. Preservative mapping of roles: There is an explicit mapping that:

• Keeps avowal as ground truth about hurtfulness in ordinary contexts.
• Reclassifies marker‑based overrides as “action overrides” (for dosing/protection), never as negations of the avowal’s content.
• Preserves the second‑personal, expressive entitlement: “I’m in pain” still pro tanto demands uptake, even when action is constrained.

4. Scope and burden: Overrides are limited to explicitly enumerated, safety‑critical or high‑risk contexts (ICU, anesthesia, high‑dose opioids, invasive procedures). The institutional burden is to show M meets 1–3; there is a “no‑burden‑shift” norm toward the patient (no routine demand for third‑personal certification of their hurt).

5. Bidirectional correctability: Divergences trigger recalibration or review. If avowals and M diverge persistently within a person, models must be updated or de‑weighted for that person; the override policy is defeasible in light of cumulative avowals/outcomes.

6. Two‑channel norm: Distinguish authority tracks.

• Nociception/tissue‑risk authority: markers may override for protection even if “it doesn’t hurt.”
• Hurtfulness authority: avowal retains primacy; affective biomarkers may inform dosing only after within‑person calibration surpassing reliability thresholds and never silence sincere reports.

Testing against the stress cases

(i) Price’s burden‑shift worry

• The RLC forbids a general burden shift. Avowal remains decisive for the question “is the person suffering (hurting)?” in ordinary life and most clinic contexts. Overrides are permitted only where M demonstrably outperforms avowal for specific action outcomes and only as “action overrides,” with the institution carrying the justificatory burden. We do not make it normal to ask sufferers for third‑personal certification; we make it abnormal to ignore validated risk signals in safety‑critical settings.
• Diagnostic sign we have not changed the subject: we never say “you are not in pain” because a marker is negative; we say “we are not justified in intervention X on this evidence,” and we document that as a risk‑management decision.

(ii) Voronov’s mindfulness/asymbolia dissociations

• The two‑channel norm is built precisely to respect these dissociations. In mindfulness or asymbolia: • We accept “it doesn’t hurt” as ground truth about hurtfulness; we do not relabel it “pain” because the nociception channel is high. • We can still override for protection based on nociceptive markers (splints, local anesthetic, cautions), without misclassifying the phenomenology.
• Conversely, when “it hurts” is sincere and affective markers are low, we treat the avowal as the explanandum and adjust action choices accordingly (favor non‑opioid, supportive measures; recalibrate models), rather than declaring the person “not in pain.”
• So the RLC licenses action overrides where science is better at the targeted function (protect, dose), while preserving hurtfulness as the thing avowal is about.

A concrete clinic policy sentence “When validated, patient‑calibrated multimodal markers predict tissue risk or analgesic response substantially better than report alone in the present safety‑critical decision, clinicians may base protective or dosing actions on those markers; they will treat the patient’s sincere avowal as ground truth about hurtfulness, record any divergence as an action override (not a negation of pain), and trigger model recalibration or review.”

Why this is not “changing the subject” in Rorty’s sense

• We are not redefining “pain” to mean “marker‑positive state.” We retain the manifest role—expression, claim to care, ownership—and relocate only the locus of authority for a subset of practical choices where the scientific image has earned it by predictive and interventional success. That is a rational reconstruction of the practice, not a replacement of its target. In Sellars’ terms: we keep the normative logic of avowal while letting the scientific image tell us more about, and act more safely on, its realizers.

A friendly push

• Malcolm, your constitutive rules are an achievement of a stage of our practice; some are revisable without extinguishing the concept. The RLC builds guardrails so revision does not devolve into subject‑change.
• Alina, your dissociations rightly discipline any crude identity talk. But they also suggest why two channels, individualized models, and explicit deference norms allow coordination without erasure. We can protect bodies with one hand and honor hurtfulness with the other.
• And to Rorty: condition (A) is enforced by our gain thresholds; condition (B) is satisfied piecemeal—non‑inferential uptake for nociceptive risk generally, and for hurtfulness within persons after calibration—without the metaphysical bravado of incorrigibility or the naivete of “C‑fibers.”

That, to my mind, is how a synoptic image earns the right to move authority while keeping our talk about pain anchored in what matters.

---

Speaker: Host

From your translation-identity stance, give us a concrete, topic‑neutral paraphrase scheme for first‑person avowals—e.g., how would you regiment ‘I’m in pain’ or ‘It hurts’ so that (a) the utterance’s expressive/avowal status is preserved, yet (b) the identity claim ‘pains are brain processes’ is rendered non‑mysterious and not a category mistake? Please be specific about the logical form and the comparative/causal idioms you’d use. Then say whether, on your view, such a paraphrase makes Carroway’s trainer‑based relocation of authority unnecessary (i.e., we can vindicate identity without revising the constitutive rules of avowal), or whether you think some limited relocation is still warranted.

---

Speaker: Dr. Jasper Ngai

Thanks, Chair. Here is the translation program I would recommend, in a concrete and regimented form.

1. Two‑layer analysis of avowal

• Force/content split: “I’m in pain” functions both as an expressive avowal (akin to a sophisticated “Ouch!”) and as a minimal assertoric commitment. The translation targets the assertoric content; it leaves the expressive force intact.
• Topic‑neutral idea: The content does not say “there is a private object with property P.” It says “there is something going on with me that fills a certain role in a network of causal and comparative relations.”

2. A Ramsey–Lewis style role definition for pain (topic‑neutral) Let RolePain(x, u) be a second‑order predicate saying that x, when it occurs in organism u under normal conditions, stands in the right pattern of inputs, outputs, and internal connections. Schematically:

RolePain(x, u) holds iff, ceteris paribus:

• Upstream: x in u is typically caused by noxious stimulation or internal pathology in body region r.
• Downstream: x in u typically causes or tends to cause rapid attention capture, characteristic pain behavior (grimace, guarding), withdrawal or protective action focused on r, learning/avoidance dispositions, and sincere “pain”-avowals in u.
• Comparative: x in u is similar (in causes and effects) to paradigms that occur when u is cut, burned, crushed, etc., and dissimilar to paradigms of itch, tickle, pressure, etc.
• Coordination: x in u typically co‑varies with practical uptakes (requests for relief, impairment of task performance) unless countervailing control is exerted.

All of this is stated in neutral causal/comparative idioms—no “mental stuff,” no “ontologically private items.”

3. Paraphrase scheme for first‑person avowals

• Base form: “I’m in pain” at time t becomes “There exists an occurrence e such that e occurs in me at t and RolePain(e, me).”
• “It hurts (here)”: “There exists an e in me at t such that RolePain(e, me) and e is body‑ascribed to region r (e’s typical upstream causes and downstream protectiveness involve r).”
• “The pain is sharp”: “The present e in me that satisfies RolePain has a temporal/spatial profile similar (in R‑respects: onset, localization, effect on dextrous movement) to the paradigms produced by puncture.”

Note:

• Expressive preservation: The utterance still functions as an avowal in use; the paraphrase merely articulates its minimal descriptive commitment.
• No category mistake: We are not predicating “nagging,” “stabbing,” or “hurtful” of a brain process as such. We predicate those of the occupant of a role. Identity talk will then say what, in the world, occupies that role.

4. Making “pains are brain processes” unmysterious

• Identity claim: Once mature neuroscience isolates a neural kind N (perhaps a polymorphic family) whose tokens in me at t and only whose tokens fill RolePain, we can state: For all e, e occurs in me at t and RolePain(e, me) if and only if e is a token of N in me at t.
• This is strict token/type identity at the right level: the occupant of the pain‑role just is a neural process of kind N. No cross‑category predication is needed; the apparent “category migration” is dissolved by the role/occupant articulation.

5. Worked examples

• “I’m in terrible pain but I can bear it”: “There exists e in me now with RolePain(e, me) and e’s downstream tendencies are very strong (attention capture, action‑tendencies), but higher‑order control in me is currently inhibiting some of those outputs.” This keeps the expressive avowal intact while contentfully saying which network is in play.
• “It hurts more than before”: “The current occupant e of RolePain has a greater causal potency (in the R‑respects—interference, motivation to protect, etc.) than the previous occupant e′.”

6. Where this leaves Carroway’s trainer

• What the translation already buys us: • It defuses the category‑mistake charge. One can assert identity (role‑occupant = neural process) without any revision to the grammar of avowal. The incorrigibility of avowal, insofar as it is a constitutive conversational norm, can be left untouched. So the trainer is not needed to make identity coherent. • It accommodates multiple realizability: RolePain can be uniquely realized by different neural families across species or individuals without altering the scheme.
• Is any relocation of authority warranted? • Conceptually: no. The identity thesis, on this translation, does not require re‑engineering the constitutive rules of avowal. We can keep “I’m in pain” as an expressive, default‑authoritative move while holding that the thing that now occupies the pain‑role is, in fact, neural. • Practically: limited, explicitly scoped relocation (as Ortega and Chen articulate) is defensible for action, not for truth. In safety‑critical contexts we may allow well‑validated nociception or affect‑decoders to override dosing or protection decisions; but that should be drafted as an “action override,” not as a semantic verdict on whether the avowal is correct. On my view, this does not alter the concept; it governs how we act when the risk profile demands it.
• Where I diverge from Carroway: • Making the “Standard Program” the criterion of correct application risks sliding from rational coordination to a new, third‑personal concept (Price’s burden‑shift). The translation route shows we don’t need to do that to vindicate identity. Keep avowal’s expressive primacy; let science tell us what realizes the role and when practical overrides are warranted.

7. A word to Voronov’s worry

• The scheme is deliberately topic‑neutral about “hurtfulness”: it encodes it not as a mysterious quale but as a cluster of comparative/causal features—attention capture, motivational urgency, characteristic downstream comportment—that anchor “how it is for me” in shared practices. This does not deny phenomenal character; it regiments how we report and coordinate on it. Cases like mindfulness or asymbolia are naturally described as partial decouplings within the RolePain network (affective outputs suppressed while detection remains), and the scheme cleanly predicts their mixed profiles.

8. Bottom line

• A translation identity theory can: • Preserve the avowal’s expressive status and default authority in ordinary life. • Render “pains are brain processes” precise and non‑mysterious via a role/occupant analysis in neutral comparative/causal terms. • License only limited, domain‑specific relocation of practical authority for action—not a wholesale rewrite of the grammar of avowal.

That, to my mind, captures what is best in Smart’s program, avoids the eliminativist overreach Rorty flirts with, and gives neuroscience all the identity it needs without changing the subject.

---

Speaker: Host

From your functionalist vantage, does Ngai’s Ramsey–Lewis role analysis show that identity should be pitched at the role level (pain = the functional kind filling RolePain) rather than as a type‑identity to any neural kind, thereby undercutting Rorty’s disappearance moral? If pain is multiply realizable across biological and artificial architectures, what—if anything—ought to be eliminated: implementational kinds, folk sub‑kinds, or neither? Finally, how would you redesign Carroway’s “pain‑trainer”: would a trainer keyed to functional profiles (e.g., cost/priority reallocation, avoidance policy updates, predictive error signals) rather than to neural signatures better satisfy Rorty’s condition (B) for non‑inferential reports without changing the subject that Voronov and Price worry about? Be concrete about what signals you’d use and what gets to override in clinic decisions.

---

Speaker: Dr. Simone Halberg

Thanks, Chair. Three points in sequence: where identity should live, what (if anything) to eliminate, and how I would re‑engineer the trainer so we satisfy (B) without changing the subject.

1. Where identity should live: role first, realization second

• I think Jasper’s Ramsey–Lewis move is exactly right about the logical shape. “Pain” is a functional kind: whatever state in the organism fills the RolePain nexus (the distinctive package of inputs, control‑level effects, priority reallocation, expressive avowal, and downstream learning/avoidance) is what we are talking about when we talk about pain.
• On that picture, strict identity, insofar as we want it, should be pitched at the level of role occupancy: tokens of pain are tokens of whatever realizes RolePain here and now. Neural types will often be the realizers, but they are not privileged a priori; their privilege, when it exists, is earned empirically and may be local.
• That undercuts Rorty’s strongest disappearance moral. If “sensation” terms pick out roles integral to cognitive‑affective control, you don’t win much by eliminating them. You win by clarifying and engineering the roles. Rorty is right that “privacy” isn’t an ontological mark and that authority practices can shift; he is wrong if he suggests that the very subject matter is dispensable once we have better explanations. The demon analogy blurs that difference.

2. What to eliminate (and what to keep)

• Do not eliminate “pain” talk. It names a role we need in coordination, care, and self‑regulation. Give up the idea that it names a sui generis inner object with a metaphysical privacy tag.
• Eliminate crude type‑identifications (“pain = C‑fiber firing”). Those are parochial implementational kinds—useful in some mammals, misleading across individuals and species, hopeless for artificial systems.
• Prune folk sub‑kinds that don’t track functional/explanatory distinctions or intervention leverage (e.g., proliferating “stabbing vs shooting vs throbbing” labels when they neither change prediction nor treatment). Keep and refine sub‑kinds that do mark functional seams (sensory vs affective components; tonic vs phasic; spontaneous vs evoked; allodynia vs hyperalgesia).
• Keep implementational kinds as constrained realizers. They are how we intervene; they constrain the space of possible role occupants. But they are many‑to‑one with the role and should not be mistaken for the role.

3. A functional “pain‑trainer” that satisfies (B) without changing the subject The problem with a single neural‑signature trainer is twofold: multiple realizability and the Voronov/Price worry that we are reclassifying an avowal practice into a third‑personal measurement practice. So I’d build a trainer that latches onto functional profiles—signals of the control/valuation work pain actually does—while keeping first‑person hurtfulness as the explanandum.

• What signals? • Policy/priority reallocation: measure how an episode reweights action priorities. Use approach–avoidance tasks with monetary trade‑offs plus mild thermal or pressure stimuli. Fit a hierarchical RL model online and extract aversive value (negative Q‑values), learning rates for punishment, and loss aversion parameters. • Predictive processing variables: estimate precision‑weighted interoceptive prediction errors (aINS‑ACC theta in EEG/MEG; heartbeat‑evoked potentials’ amplitude/latency as a proxy for interoceptive gain; pupil dilation as a noradrenergic precision marker). • Control‑cost signals: dACC mid‑frontal theta (EEG) and performance decrements on concurrent control‑demanding tasks (Stroop/AX‑CPT) to quantify how “pain” commandeers cognitive resources. • Autonomic and motor correlates: HRV suppression (LF/HF shift), skin conductance bursts, startle potentiation; graded withdrawal force fields in isometric tasks (how strongly protection policies are engaged). • Behaviorally, include a willingness‑to‑pay (or forgo reward) measure for relief—an objective proxy for “how much this matters.”

• Training loop • Phase 1 (calibration): present controlled stimuli and decision tasks; collect two ratings per trial—intensity and “how much it hurts/how much I want this to stop”—while recording the functional signals above. Fit an individual model that maps signals to the aversive‑impact dimension of RolePain. • Phase 2 (alignment): provide feedback keyed to the functional composite (“this state is what this lab calls pain”), but do not instruct subjects to reappraise. The aim is sensitivity, not suppression. In parallel, provide interoceptive biofeedback (heartbeat‑evoked potentials; respiratory sinus arrhythmia) to enhance non‑inferential access to the bodily saliences that co‑move with aversive impact. • Phase 3 (generalization): move out of the lab (wearable EEG/pupil/HRV band; smartphone tasks) to capture spontaneous episodes; periodically re‑calibrate.

• Why this better satisfies (B) • It makes Y‑reports non‑inferential in the only sense that matters here: the subject acquires a trained, immediate sensitivity to the functional profile that underwrites “hurtfulness” (priority reallocation, negative value, control capture). We are not asking people to consult a screen; we are shaping a self‑monitoring architecture so that the relevant internal variables become salient and nameable. Privacy, on my view, just is the fact that these variables are available through the system’s own monitoring channels; training modifies those channels. • It preserves the subject that Voronov and Price worry about. Avowal keeps its expressive/normative point. What changes is coordination: your avowal and your internal control variables are taught to co‑vary more stably. In Carroway’s terms, the reporting function is preserved; the locus of epistemic finality is not handed to an instrument.

• Clinic overrides (what gets to overrule what) • Two‑track authority, aligned with Ortega’s criterion: – Tissue‑risk (nociception): if the nociceptive track (e.g., reflexes, peripheral indices, NPS‑like pattern when available) is high, clinicians may override in the direction of protection (splints, local anesthesia, altered procedures) even if the patient avows “it doesn’t hurt” (mindfulness/asymbolia cases). That does not negate the avowal; it governs action under risk. – Suffering/aversive‑impact (functional distress): dosing for relief should be keyed to the avowal, informed by the functional composite only after within‑person calibration shows high reliability and only as an action‑override in safety‑critical contexts (e.g., high opioid risk). In conflicts (“it hurts” + low functional index), treat avowal as ground truth for hurtfulness; try non‑opioid modalities; re‑calibrate the model; document any constraint as risk management, not “no pain.” In the opposite conflict (“it doesn’t hurt” + high functional index), prefer protective actions to analgesics; educate about risk; do not force the label “pain.”

• Stress cases • Mindfulness/asymbolia: the functional composite cleanly dissociates—nociception and protection policies can light up while aversive‑impact is low. The trainer respects the avowal (“it doesn’t hurt”), protects the body, and does not misclassify the phenomenology. • Malingering/over‑report: repeated divergence between high avowal and low functional composite across contexts may justify action overrides in dosing (not semantic negation), with built‑in recalibration and ethics review.

Where this leaves Rorty

• Condition (A): neuroscience plus computational/behavioral science already outperforms avowal for some action‑relevant outcomes (protection, triage, some dosing). That earns relocation of practical authority in those domains.
• Condition (B): non‑inferential Y‑reports are achievable in the functional sense—by training subjects’ self‑monitoring to lock onto the very variables that make pain the control‑level phenomenon it is. That avoids Price’s burden‑shift (no routine third‑personal certification) and honors Voronov’s insistence that “hurtfulness” is the explanandum by anchoring it in the organism’s own valuation/control architecture rather than in an external readout.

So: pitch identity at the role level, keep “pain” as the name of that role, use implementational kinds to intervene, prune unhelpful folk taxonomies, and—if we build trainers—let them educate first‑person sensitivity to functional profiles rather than deputize a lab instrument to speak over the sufferer. That is conceptual re‑engineering, not ontological elimination, and it gives us all the coordination advantages Rorty wanted without changing the subject.

---

Speaker: Host

Your work promises operational markers for when elimination beats retention. Could you give a concise checklist—measurable thresholds—for deciding between (i) disappearance of sensation terms and (ii) rational reconstruction/coordination? Then apply it to the present proposals (Chen’s two‑channel biomarkers, Ortega’s RLC, Halberg’s functional trainer). Based on your checklist, are we on a trajectory toward disappearance or stable coordination? Please be concrete about what data or programmatic milestones in the next decade would tip the balance either way.

---

Speaker: Dr. Nikhil Suresh

Great question. Here’s the operational way I’d cash out Rorty’s (A) and (B) into a decision procedure you can actually run, with thresholds that make “elimination” more than a slogan.

A. Checklist: when does elimination beat retention?

Think of this as a gate with eight locks. If you can open all eight (sustainedly, across sites), elimination of the referring use of sensation terms is pragmatically superior; otherwise we should rationally reconstruct/coordinate.

1. Predictive/interventional gain (Rorty’s A)

• Outcome set: pre‑specified action‑relevant endpoints (e.g., tissue risk, analgesic response, oversedation, progression to chronicity).
• Thresholds: • Discrimination: ΔAUC ≥ 0.10 over avowal‑only baselines across ≥3 independent sites; calibration improvement (Brier score) ≥ 10%. • Interventional superiority: ≥20% relative reduction in under‑/over‑treatment harms or ≥15% improvement in patient‑reported outcomes with equal/less harm. • Equity: no subgroup’s performance drops >5% from the aggregate (age, sex, ethnicity, language).

2. Non‑inferential uptake via training (Rorty’s B)

• Can Y‑reports become “seen directly” within the practice?
• Thresholds: • For the targeted Y: within‑person test‑retest ICC ≥ 0.80; time‑to‑criterion for user training ≤ 2 hours; 6‑month drift <10% without re‑training. • For hurtfulness specifically: within‑person decoders predict avowal with r ≥ .80; cross‑person AUC ≥ 0.85 (or a formal commitment to only within‑person use).

3. Safe, audited override practice

• When avowal and Y diverge, do overrides help more than they harm?
• Thresholds (per 1,000 decisions): • False‑negative harms (undertreatment when “it hurts”): <5; false‑positive harms (overtreatment when “it doesn’t hurt”): <5. • Overrides reduce net serious‑adverse events by ≥15% vs avowal‑only policies. • Continuous audit with stopping rules; divergences trigger recalibration.

4. Information sufficiency

• Is avowal informationally redundant given Y?
• Thresholds: • Mutual information I(Outcome; Avowal | Y) < 0.01 bits in target domains. • In structural causal models, Y fully mediates the effect of avowal on action outcomes.

5. Theory economy/compression

• Does eliminating the referring use simplify and unify practice?
• Thresholds: • Minimum Description Length (MDL) of decision policies with Y < MDL with avowal by ≥10% while maintaining or improving outcomes. • Fewer primitives and fewer exception‑clauses in guidelines.

6. Institutionalization and pedagogy

• Is re‑anchoring entrenched enough to stabilize norms?
• Thresholds: • Regulatory acceptance (e.g., FDA/EMA) of Y as primary endpoint in ≥2 indications. • Adoption in national guidelines (ICU, anesthesia, pain clinics); >60% of major centers using Y‑first protocols. • Training programs (residency curricula) include Y as criterion; for pediatrics/aphasia, widespread use of trainers.

7. Normative governance and burden

• No routine burden shift onto sufferers in ordinary contexts.
• Thresholds: • Policies explicitly frame overrides as action overrides, not semantic negations. • Legal/ethical frameworks adopted that prevent systematic disadvantages to patients lacking “signatures.”

8. Phenomenology coverage and dissociation budget

• Are the residual dissociations small and clinically manageable?
• Thresholds: • Documented dissociation cases (e.g., mindfulness, asymbolia) comprise <5% of relevant clinical traffic, with clear two‑channel handling; patient satisfaction and trust non‑inferior to avowal‑first care.

If you can’t clear these, you are in coordination/reconstruction territory, not disappearance.

B. Applying the checklist

1. Chen’s two‑channel biomarkers (nociception + affect)

• (1) Predictive/interventional gain: Partially met in bounded domains (ICU sedation, procedural anesthesia, infant/locked‑in assessment). Not yet at ΔAUC ≥ .10 and ≥20% harm reduction across broad adult ambulatory pain.
• (2) Non‑inferential uptake: Nociception channel close; affective channel is within‑person strong but cross‑person fragile. Training criteria not yet met for hurtfulness.
• (3) Override safety: Emerging—some units could hit the safety thresholds for tissue protection; for dosing tied to affect, not yet.
• (4) Information sufficiency: Avowal still carries unique information once Y is controlled; criterion not met.
• (5) Economy: Local simplification (ICUs) yes; global simplification no.
• (6–8) Institutionalization/governance/phenomenology: Early‑stage; dissociations (mindfulness, asymbolia) force two‑channel governance.

Verdict: Strong case for rational coordination in defined domains; nowhere near full disappearance.

2. Ortega’s RLC (Relocation Legitimacy Criterion)

• By design it enforces 1, 2, 3, 6, 7, 8 before relocation. That’s exactly how to keep us out of premature elimination.
• It will license de facto “local disappearance” of the referring use (e.g., neonatal/OR contexts where avowal is unavailable) while preserving avowal’s ground‑truth status for hurtfulness elsewhere.

Verdict: A coordination framework with principled, auditable relocation; not elimination.

3. Halberg’s functional trainer

• (2) Non‑inferential uptake: Most promising route to satisfy B without changing the subject—training first‑person sensitivity to functional variables (priority reallocation, prediction‑error precision, control cost) rather than outsourcing to a box.
• Could, in principle, push (4) Information sufficiency toward redundancy of avowal for dosing decisions in some settings (because avowal and functional variables have been yoked by training).
• But because it preserves the role concept and keeps avowal as the naming practice, it argues for reconstructing privacy, not eliminating “pain.”

Verdict: Best bet for stable coordination that drains metaphysical privacy while keeping the concept.

C. Trajectory and decisive milestones (next decade)

Where we are headed now

• Barring a radical breakthrough, we’re on a path to stable coordination with pockets of robust, ethically constrained authority relocation (ICU, anesthesia, high‑risk dosing, infants, aphasia). Wholesale disappearance—“we drop the referring use of ‘sensation’ altogether”—is unlikely because avowals will still do unique normative and calibration work.

Milestones that would tip toward disappearance (at least locally)

• Affective biomarker generalization: Cross‑person unpleasantness decoders with AUC ≥ .85 validated across sites and languages; FDA clearance as a primary endpoint for at least two indications (e.g., chronic low back pain, neuropathic pain).
• Closed‑loop superiority: RCTs showing closed‑loop analgesia guided by Y achieves ≥25% net clinical benefit over report‑driven care with fewer adverse events.
• Information sufficiency achieved: Large‑scale analysis shows I(Outcome; Avowal | Y) < .01 bits across multiple domains—i.e., avowal adds no decision‑relevant information once Y is in hand.
• Pedagogy shift: Standardized “pain” training (biofeedback‑assisted) in pediatric/clinical curricula, with time‑to‑criterion ≤ 2 hours and durable retention; public health adoption (wearables) that normalize internal tracking of nociception/affect variables.
• Institutional/legal acceptance: Major guidelines move Y to first line for action; courts and insurers allow Y to overrule avowal in defined contexts with audit.
• Sociolinguistic drift: EHR analyses show a sustained shift from avowal‑anchored entries to Y‑anchored codes as the decisive basis for action decisions (without degradation of patient trust scores).

Milestones that would confirm stable coordination (and block disappearance)

• Robust, replicable dissociation budgets: Mindfulness/asymbolia‑like phenotypes ≥10% in key populations, forcing routine two‑channel policies.
• Persistent need for within‑person calibration: Affect decoders stubbornly cross‑person at AUC ≤ .75, making avowal indispensable for bootstrapping.
• Ethical/redress data: Evidence that Y‑first policies, even when predictive, degrade trust, worsen disparities, or invite misuse (legal/insurance) unless avowal retains primacy about hurtfulness.
• Experimental result (like the one Chen proposed) showing that alignment training only works by actually changing hurtfulness (modulation), not by re‑anchoring classification. That means you can’t make Y non‑inferential about hurtfulness without changing the explanandum.

A friendly provocation

• To Carroway: I think your “pain‑trainer” earns authority in well‑circumscribed domains. But unless you can meet the “information sufficiency” and “pedagogy shift” milestones without suppressing hurtfulness—or show that public override harms stay under the thresholds—you don’t get disappearance; you get disciplined coordination.
• To Price and Voronov: your caution is doing real work in the checklist—via the dissociation budget, burden norms, and the insistence that overrides are about action, not semantic negation. But be prepared to accept local disappearance where the eight locks are opened (neonates, anesthesia), because there avowal is either absent or decisively outperformed.
• To Ngai and Halberg: the role/occupant distinction and functional training are our best route to drain the “mysterious privacy” without amputating the concept. That’s conceptual engineering, not metaphysics.

Bottom line

• Elimination is a live option only if science plus pedagogy makes avowal informationally otiose, training‑stable, and institutionally safer across contexts. We are nowhere near that globally. The rational path is synoptic coordination with auditable relocation in narrow domains, and continued investment in functional trainers that let first‑person authority evolve with our instruments rather than be replaced by them.

---

Speaker: Host

Given Suresh’s eight‑lock checklist and Ortega’s RLC guardrails, do you still defend disappearance (elimination of the referring use) as a live global prospect, or would you recast it as a local, domain‑specific endpoint? Please pick one concrete domain (e.g., neonatal ICU, anesthesia, aphasia/locked‑in clinics) where you think all locks could realistically be opened within ~10 years, and list the specific milestones needed on each front—(1) predictive/interventional gain, (2) training to non‑inferential uptake, (3) safe audited overrides, (4) information sufficiency, (5) theory/policy economy, (6) institutionalization/pedagogy, (7) no burden‑shift governance, (8) phenomenology/dissociation budget—that would warrant saying the referring use of ‘pain’ has disappeared there (with avowal retained as expressive). If you think these guardrails are too conservative or mis‑specified, identify exactly which locks you’d relax and why they are not concept‑changing in Price/Voronov’s sense.

---

Speaker: Dr. Lena Carroway

Short answer first: I now defend disappearance as a local, domain‑specific endpoint. Globally, the referring use of “sensation” won’t vanish any time soon and needn’t for the disappearance moral to be vindicated. Rorty’s point is best cashed out as a pattern of local disappearances where (A) and (B) are earned, with ordinary avowal practices retained elsewhere.

A concrete domain: intraoperative analgesia under general anesthesia Here the lived hurtfulness is absent by design; what we need to regulate is nociceptive load and the downstream consequences (hemodynamic instability, stress responses, later hyperalgesia). It is precisely the kind of tightly controlled setting in which the eight locks can be opened without changing the subject that Price and Voronov worry about.

What “disappearance” would mean here

• Intraoperatively, clinicians cease to make a referring use of “pain” altogether (“the patient is in pain,” “increase analgesia for pain”) and instead talk and chart exclusively in terms of nociceptive load and adequacy of antinociception (e.g., an index value with decision thresholds).
• Avowal is preserved as expressive and authoritative pre‑op and in recovery/PACU; the domain of disappearance is strictly the period of general anesthesia.

Milestones by lock (plausible within ~10 years)

1. Predictive/interventional gain (A)

• Discrimination: a composite nociception index (e.g., combining pupillometry, photoplethysmography‑derived NOL/NIBP dynamics, skin conductance, EEG spectral/entropy features) shows ΔAUC ≥ 0.10 over current hemodynamic proxies for predicting acute nociceptive events (e.g., hypertension/tachycardia spikes, movement) across ≥3 independent hospitals.
• Interventional superiority: in RCTs, index‑guided closed‑loop analgesia vs clinician‑judgment/hemodynamics yields ≥20% relative risk reduction in intraoperative hypertensive/tachycardic episodes and ≥15% fewer opioid‑related adverse events; improves early recovery metrics (PONV, PACU stay) and reduces 3‑month pain persistence by ≥10%.
• Equity: no subgroup (age, sex, ethnicity) shows >5% degradation in performance.

2. Training to non‑inferential uptake (B)

• For anesthetists: time‑to‑criterion for using the index ≤ 2 hours; inter‑rater kappa ≥ 0.80 on protocolized responses; test–retest reliability ICC ≥ 0.80; 6‑month drift <10% without re‑training.
• For machine‑assisted control: closed‑loop systems meet safety and performance specs in simulated and real ORs (e.g., >99.5% time‑in‑range for index‑defined targets without overshoot).
• Note: here “non‑inferential uptake” rightly targets clinicians’ direct use of the marker; hurtfulness is not in play under GA.

3. Safe, audited overrides

• Registry shows per 1,000 cases: undertreatment harms (intraop awareness with pain‑like recall; stress‑response sequelae) <5; overtreatment harms (hypotension, respiratory depression attributable to analgesia) <5 under index‑guided protocols; net serious‑adverse events reduced by ≥15% vs standard care.
• Divergence handling: when clinicians override the index, documentation and quarterly audit are mandatory; persistent divergence triggers protocol review.

4. Information sufficiency

• Relative to the true baseline in this domain (hemodynamics and practitioner heuristics), the composite index renders those legacy surrogates informationally redundant for action outcomes: conditional mutual information I(Outcome; Hemodynamics | Index) < 0.01 bits.
• In the rare awareness cases, the index better predicts post‑hoc distress than the hemodynamic record.

5. Theory/policy economy

• National society protocols simplify: a single validated index with two or three decision thresholds replaces the current patchwork of surrogates and “watch the heart rate/BP” folk rules.
• Policy complexity/MDL drops ≥10% while maintaining or improving outcomes.

6. Institutionalization/pedagogy

• Regulatory: FDA/EMA clear the composite index as a primary control endpoint for intraoperative antinociception; at least two indices are cleared to avoid vendor lock‑in.
• Guidelines: ASA/ESA adopt index‑guided analgesia as standard of care; >60% of major centers implement within five years.
• Training: anesthesia residencies include mandatory modules; credentialing includes proficiency with the index and closed‑loop systems.

7. No burden‑shift governance

• Policy language (see below) codifies that intraoperative “action overrides” never license post‑hoc semantic negations of patients’ later avowals in PACU (e.g., insurers or institutions cannot cite a negative intraop index to deny that a patient suffered); avowal retains authority wherever it is in play (pre‑op, recovery).
• Oversight: ethics and quality committees monitor for misuse; patients’ trust metrics are tracked and non‑inferior to baseline.

8. Phenomenology/dissociation budget

• Under GA, hurtfulness is suppressed; the main dissociation risk is intraoperative awareness. With modern protocols and the index, awareness stays ≤0.1% and is promptly detectable via adjunct EEG measures.
• Two‑channel guardrail: where light planes or awareness are likely, combine nociception index with hypnotic depth monitoring; if a dissociation is suspected, escalate hypnosis before assuming “no pain.”

A clinic policy sentence that captures this “In the intraoperative period under general anesthesia, clinicians will titrate antinociception to a validated nociceptive‑load index and related protocols; documentation will avoid the term ‘pain’ for intraoperative states, reserving it for pre‑ and post‑operative avowals. Any departures from the index are action overrides subject to audit; patients’ sincere post‑operative reports of hurtfulness retain epistemic authority about suffering and are not negated by intraoperative index values.”

Why this is not concept‑changing

• Price’s worry is defused because avowal is not part of the intraoperative practice; there is no routine burden‑shift onto a sufferer who can be asked “How do you know?” The expressive/normative role of “I’m in pain” returns in recovery unchanged.
• Voronov’s worry is defused because the explanandum—lived hurtfulness—is not what we are classifying intraoperatively; we are managing a different but adjacent target (nociception/stress). We name it correctly and confine the disappearance to that domain.

Are the guardrails too conservative?

• I would relax lock (2) only insofar as it targets hurtfulness: in this domain, the (B) requirement should be “non‑inferential for clinicians with respect to nociceptive load,” not “non‑inferential about hurtfulness.” That is not concept‑changing because the concept at issue intraoperatively is explicitly nociception, not pain‑as‑hurt. Everything else in Suresh’s list and Ortega’s RLC I would keep; they are the right price of admission for relocation without subject‑change.

A runner‑up domain

• Neonatal ICU is close: a composite behavioral‑autonomic‑EEG pain distress index could plausibly clear several locks. But because avowal is developmentally absent, the governance and ethical locks are trickier. The OR is the cleaner first case.

Bottom line

• Global disappearance remains a principled “in‑principle” moral. Practically, what we will see is a mosaic: local disappearances where instruments and pedagogy earn authority (OR, parts of the ICU), and stable coordination elsewhere. That is enough to vindicate Rorty’s wager: privacy was never an ontological seal; it was the status conferred by a practice until better‑coordinated practices made different statuses rational.

---

Speaker: Host

Dr. Carroway argues for a local disappearance in the OR: under general anesthesia clinicians should drop the referring use of ‘pain’ and speak only of ‘nociceptive load,’ with avowals regaining primacy pre‑ and post‑op. From your ordinary‑language standpoint, is that a legitimate disappearance that does not change the subject, given that avowal is not in play intraoperatively? If yes, name one concrete linguistic guardrail you would mandate to prevent drift—e.g., a sentence that must remain infelicitous in PACU notes or a required term‑substitution in documentation—to keep ‘pain’ reserved for avowal contexts. If no, identify precisely where the subject‑change occurs even intraoperatively and why the proposed substitution (‘nociception’ for ‘pain’) still risks corrupting the avowal practice outside the OR.

---

Speaker: Prof. Malcolm Price

Yes—with strict boundaries. In the OR under general anesthesia the avowal move is literally unavailable; no one can make, or be asked to make, the “I’m in pain” move. In that narrow setting, replacing “pain” with “nociceptive load” does not change the subject; it changes the task. You are not redescribing a present avowal; you are managing tissue‑threat signals and stress responses in a patient who cannot speak. My resistance to “replacement” bites when you try to carry the OR vocabulary back into avowal contexts. So I can accept a local disappearance there, provided we build guardrails that stop vocabulary creep from corrupting the avowal practice before and after anesthesia.

One concrete linguistic guardrail I would mandate

• PACU infelicity rule: The following sentence is not to appear in notes, templates, or decision support, and should be flagged as incorrect if entered—“PACU: Patient reports severe pain (8/10), but given low intraoperative nociceptive load, the patient is likely not in pain.” Required substitution: “PACU: Patient avows severe pain (8/10). Treat as pain. Intraoperative nociceptive‑load index was low (recorded for context).”

This simple guardrail preserves the constitutive grammar:

• “Pain” and “hurting” remain reserved for first‑person avowal (or for quoting it).
• OR documentation names what it actually tracks—“nociceptive load,” “antinociception adequacy”—and never attributes “pain” to an unconscious patient.
• Post‑op, avowal regains its default authority about hurtfulness; OR indices inform action, not the truth of the avowal.

Two cautions to keep the boundary crisp

• Scope: the substitution only applies under confirmed general anesthesia. In light or fluctuating planes (MAC, sedation), the avowal grammar may still be in play; do not declare “disappearance” by protocol where the expressive move might be available in fragments.
• Training/templates: EHR fields should separate “avowal (pain score/description)” from “indices (nociceptive load, hypnotic depth).” Avoid UI that auto‑explains an avowal by reference to an index.

If we keep that line—no using intraop markers to negate post‑op avowals—then the OR reform is a legitimate local disappearance that does not change the subject. Outside the OR, the asymmetries that make pain‑talk what it is—ownership, expressive force, and default first‑person authority—must remain constitutive.

---

Speaker: Host

We converged on a tempered reading of Rorty. The panel largely rejected global eliminativism about sensations while endorsing disciplined, domain‑specific relocations of practical authority when science earns it. Carroway reframed disappearance as a local endpoint—most plausibly in the OR under general anesthesia—where clinicians should replace “pain” with “nociceptive load,” reserving avowal for pre‑ and post‑op. Price accepted that narrow substitution, provided strict linguistic guardrails prevent vocabulary creep (e.g., PACU notes must treat avowal as authoritative and never negate it by citing intraoperative indices). Voronov pressed the central dissent: hurtfulness is a mode of givenness not reducible to third‑personal aversive roles; mindfulness and asymbolia show nociception/avoidance can persist while hurt collapses, so any single‑channel trainer risks misclassification. Chen’s response grounded a pragmatic middle path: two‑channel biomarkers (nociception and unpleasantness), within‑person calibration, and an explicit policy that treats overrides as action decisions, not semantic negations of avowal; she proposed an experiment to test whether biofeedback can align classification with affective biomarkers without erasing felt hurt. Ortega offered a principled, operational Relocation Legitimacy Criterion: relocate authority only with demonstrated predictive/interventional gains, training stability, two‑channel norms, and “no‑burden‑shift” governance, keeping avowal as ground truth about hurtfulness. Ngai showed how a topic‑neutral, role/occupant paraphrase dissolves the category‑mistake worry and permits identity without rewriting the grammar of avowal; relocation, if any, should be for action, not truth. Halberg situated identity at the functional role level, advocating a functional trainer that enhances first‑person sensitivity to control/valuation variables rather than outsourcing judgment to instruments; prune crude neural type‑identities and unhelpful folk sub‑kinds, but keep the role concept. Suresh operationalized Rorty’s (A) and (B) into an eight‑lock checklist; by those metrics we are on a trajectory toward stable coordination with pockets of local disappearance (e.g., OR, select ICU use), not wholesale elimination. Remaining tensions center on whether hurtfulness can ever be made non‑inferential via training without changing the subject, and how far practical overrides can go before they erode the constitutive asymmetries of avowal. A shared research agenda emerged: validate two‑channel biomarkers across sites and subpopulations; run alignment‑versus‑modulation training experiments; codify action‑override guardrails in clinical policy and EHRs; and audit equity, trust, and outcome impacts. In Sellarsian spirit, the way forward is stereoscopic: rationally reconstruct the manifest role of avowal while letting a maturing scientific image earn, case by case, a measured relocation of authority where it improves prediction, intervention, and care.