SSalarsNet

AI · Article 43 of 64

Pain, Grief, Guilt and the Dangerous Dream of Eliminating Every Unpleasant State

Should every unpleasant internal state be treated as a technical defect?

By Randy Salars · October 7, 2026

Consider a hypothetical woman who cannot feel physical pain and walks into a hospital with a broken ankle she did not notice. She has stepped wrong a week earlier, felt nothing, continued walking, and now the joint is badly damaged. This is what life looks like when a warning signal is missing. The initial injury may have been an ordinary accident. The absence of pain let her continue loading the damaged joint.

That case is easy to accept as a caution. The harder case is the one where the signal arrives and we want it gone: the grief after a death, the guilt after a betrayal, the anger after an injustice, the dread before a genuine danger. Here the temptation is to read the discomfort itself as the malfunction—to treat the unpleasantness as a bug to be patched, whether with a drug, an algorithm, or a well-intentioned product feature. Sometimes that reading is right. Sometimes the state is a disorder that deserves treatment. The question is how to tell the difference before you silence the messenger.

The question hiding inside “symptom relief”

Every intervention against a negative state encodes an implicit diagnosis. When a system decides that an emotion should be reduced, it has already ruled on three things: that the state is disproportionate to the situation, that reducing it will leave the person better off, and that nothing important was being carried by it. Those are empirical and moral claims, and they are frequently made without being stated.

The practice of medicine already has a word for getting this wrong, and it is older than any of our current technology. It is the distinction between treating a symptom and treating a cause. A cough suppressant is the right move for a cough that exists to clear an irritation and has already done its work; it is the wrong move for a cough caused by a tumor. The drug does not know the difference. Someone has to.

What has changed is the scale and intimacy of the intervention. A person can now carry a device that infers internal state from voice, sleep, typing rhythm, heart-rate variability, or implanted electrodes, and that acts on that inference—coaching, nudging, or, in closed-loop neurostimulation, directly modulating activity. The inference is fast and the action can be automatic. That combination makes the diagnostic question urgent, because an automatic system that suppresses first and asks later will be efficient and wrong in exactly the cases that matter most.

Pain as a protective system

The clearest evidence that some unpleasant states are functional comes from people who never have them.

Congenital insensitivity to pain is an extremely rare condition in which a person, from birth, cannot perceive pain of any kind. It is caused by mutations in a small number of genes affecting nociceptors—the specialized neurons that detect actual or potential tissue damage. Annette Lischka and colleagues, writing in Nature Reviews Disease Primers in 2022, describe the resulting picture with bluntness: recurrent injuries, burns, and poorly healing wounds are the hallmarks of the disorder. The neurons that would have signaled damage are absent or nonfunctional, so damage accumulates unnoticed. Charcot joints—progressive destruction of a joint from repeated unperceived injury—are common. So are corneal injuries, because the reflex that would close the eye to protect it never fires.

A 2020 review by Ichrak Drissi and colleagues in the British Medical Bulletin adds a detail that matters for the argument about character. Affected children can learn to say “ow,” because they observe the social behavior around injury. What they do not learn is pain-avoidance behavior—the visceral caution that keeps a person from putting weight on a damaged limb or a hand on a hot surface. As a consequence, and particularly among boys, the failure to develop this kind of learned avoidance carries significant risk of death. Pain, in other words, does two things: it reports damage, and it teaches. Remove it entirely and both functions disappear, and the outcome is not a life without suffering but a life with more of it, arriving late and through the body’s structural failures.

The implication is not that pain is good. The implication is that pain is about something, and the “about” is why suppressing it wholesale is a mistake. The goal of analgesia should be to keep the signal that protects while removing the signal that no longer serves—which is precisely what the best modern pain medicine attempts, and precisely what a system that simply lowers a dial cannot do.

When grief stops being adaptive

Grief is the state most often mistaken for a defect, and the clinical literature here is instructive because it took decades of argument to get the boundaries right.

The psychiatric establishment added prolonged grief disorder to the DSM-5-TR in 2022, after a long and contentious review. The disorder is defined narrowly on purpose. It requires the death of a close person at least twelve months earlier, and a persistent grief response marked by intense yearning for the deceased or preoccupation with them, together with at least three of eight specified symptoms—identity disruption, disbelief, avoidance of reminders, intense emotional pain, difficulty reintegrating into life, emotional numbness, a sense that life is meaningless, intense loneliness—occurring nearly every day for at least the past month, causing clinically significant impairment, and clearly exceeding what the person’s culture and context would expect.

The narrowness is the point. Ordinary grief is not a disorder and does not need treatment. Holly Prigerson and colleagues, validating the new criteria in World Psychiatry in 2021, reported that prolonged grief disorder is statistically distinct from major depression and post-traumatic stress disorder, that it is unusually stable over time, and that it is associated with suicidal ideation and functional impairment. It affects an estimated 4 to 15 percent of bereaved adults, a minority but not a small one. And it responds to something specific. As Prigerson, Shear, and Reynolds summarized in JAMA Psychiatry in 2022, across three randomized trials with 641 participants, a grief-targeted psychotherapy produced response rates of about 71 percent, against roughly 44 percent for treatments that work on depression. In the largest trial, the antidepressant citalopram did not outperform placebo in resolving grief symptoms.

That result is a small landmark for the argument of this article. It shows that the “diagnosis before suppression” rule is not mere philosophical hand-wringing; it is good clinical practice with measurable consequences. Treating prolonged grief as if it were depression helps less. Treating it as grief, with a therapy designed for grief, helps more. The state needed to be understood before it was reduced.

The other half of the lesson is that this disorder, once identified, does deserve treatment. A person trapped for years in yearning that will not resolve is suffering needlessly, and the impulse to say “grief should be honored, never treated” is its own kind of failure—one that leaves people alone with a treatable condition. The correct posture is neither “suppress all grief” nor “sanctify all grief,” but a careful distinction that only someone who knows the person’s situation can draw.

Guilt versus shame: one repairs, the other corrodes

Guilt is the state whose function is most often overlooked, because popular culture treats it as a poison. The psychological evidence is more interesting.

Cynthia Cryder and colleagues, in a 2012 paper in Personality and Social Psychology Bulletin, ran five experiments on what guilt actually does. Their finding was that guilt is not a diffuse bad feeling that makes people generally nicer; it is a targeted signal that makes people more generous specifically toward the person they wronged, and specifically when that person can see the gesture. Guilt, on this account, functions as an accounting device that identifies a specific social debt and prompts payment of it. That is a strikingly useful design.

Crucially, guilt differs from shame in a way that matters enormously for anyone building a system to manage states. Guilt attaches to an act—“I did a bad thing”—and motivates repair of the act. Shame attaches to the self—“I am bad”—and tends to motivate withdrawal, concealment, and self-protection, because a defective self cannot be repaired by an apology. Research in children has found a similar pattern: guilt following a transgression motivates reparative behavior, and the repair in turn relieves the guilt, while shame does not follow this path.

The cross-cultural evidence has grown stronger. A 2025 registered report in Nature Human Behaviour by Catherine Molho and colleagues ran a dictator-game experiment with 7,978 participants in twenty countries, manipulating what people knew about the consequences of their decisions and whether those decisions were observable. Prosociality rose when people were informed of the harm their choices caused, and it did not rise much from being watched. The guilt mechanism—knowing you have caused harm—traveled across very different cultural settings. Being observed, the shame mechanism, mattered far less.

If guilt is a signal that prompts repair, then a technology that reliably blunts guilt has a predictable cost. It would not make people crueler in a dramatic way. It would make them slightly less likely to apologize, to make amends, to notice the accumulating small debts that relationships run on. The harm would be distributed, subtle, and easy to miss, which is exactly why a system optimizing for “user comfort” would produce it without any sense of having done wrong.

Moral injury: when the signal points at something real

There is a class of painful states that cannot be resolved by changing the person at all, because the person’s distress is a correct response to something that should distress them.

Brett Litz and colleagues gave this problem its clearest formulation in a 2009 paper in Clinical Psychology Review on moral injury in war veterans. They described potentially morally injurious events as perpetrating, failing to prevent, or bearing witness to acts that transgress deeply held moral beliefs and expectations, and argued that the resulting harm can be long-lasting in emotional, psychological, behavioral, spiritual, and social terms. They were explicit that they were neither seeking to create a new diagnosis nor to medicalize moral distress.

The clinical implication is unusual. The treatment they sketched includes elements like dialogue with a benevolent moral authority and a step devoted to reparation and forgiveness—interventions that address the person’s relationship to a real moral fact, rather than attempting to extinguish a symptom. If your distress is telling you that you did something wrong, the way through is often to reckon with having done it, and possibly to make it right. A regimen that tranquilized the distress without touching the deed would leave the ledger unbalanced, and many people in that position describe the resulting calm as a kind of exile.

This is where the analogy to physical pain becomes exact. A patient with a broken leg needs the fracture set, not just the ache dulled. A person carrying guilt over a genuine transgression may need restitution and acknowledgment, not relief. And here the risk of a pleasure-optimizing system is at its sharpest: it would be most effective precisely where it should be least applied.

Adapting people to bad conditions

The most serious version of this worry concerns society rather than individuals.

If unpleasant states are treated as defects, then the efficient response to a signal is to remove the signal rather than the condition that produced it. The evidence on hedonic adaptation makes this more than a rhetorical possibility. Ed Diener, Richard Lucas, and Christie Scollon, writing in American Psychologist in 2006, reviewed the long-standing “hedonic treadmill” model—the idea that people quickly return to a baseline level of happiness regardless of what happens to them—and found that it needed substantial revision. Set points are not neutral; most people are above neutral most of the time. People differ in their baselines. Well-being has multiple components that can move in different directions. Most importantly for our purposes, set points can change, and people differ in how much they adapt.

The discomforting corollary is that humans are already good at adapting to conditions we would not choose. If a workplace is dehumanizing, a family situation is corrosive, or a city is designed around a commute that ruins people’s health, the people inside those conditions will, given time, normalize them and largely stop complaining. That norm is partly a mercy. It is also, from the standpoint of anyone who wants the conditions changed, a problem.

A technology that accelerates and smooths that adaptation—by quieting the low-grade anger, the fatigue, the dread, the sense of futility—would make a bad situation feel acceptable faster. It would not need to be coercive to do this. It would only need to be pleasant. The signal that would have prompted someone to quit, organize, complain, or leave would be quietly doing less work. And because the state was unpleasant, removing it would look like a kindness at every step.

Nobody is proposing this as a policy. The realistic version is milder and more plausible: a suite of consumer and workplace products, each defensible in isolation, whose aggregate effect is to lower the volume on the signals that historically drove people to demand change. That is the mechanism worth naming, because it operates without any villain.

Serious relief, seriously defended

It would be a mistake to end on that caution, because the mirror-image failure is just as real, and arguably more common in practice.

There is a long tradition of insisting that suffering is ennobling and that relief is a form of cheating—that the depressed should summon discipline, that the anxious should simply try harder, that grief should be borne without help. This tradition has caused a great deal of preventable misery. Major depression is not a message from the body about a correctable circumstance; it is a disorder with serious morbidity and mortality, and effective treatments exist. Anxiety disorders, chronic pain, insomnia, and prolonged grief are all conditions in which the mechanism has gone wrong and the signal is no longer carrying useful information. Withholding treatment in the name of authenticity is not respect for the person; it is neglect dressed as philosophy.

The evidence for precise, well-targeted relief is genuinely strong, and it should be stated as strongly as the risks. A person whose panic attacks are treated is not less themselves—they are more able to be themselves. The woman who cannot sleep is not learning something valuable by lying awake. The intent of the technology matters less than whether the state it targets is still doing useful work. Where the state is a malfunction, removing it is straightforwardly good.

The interesting engineering question, then, is whether a system can help distinguish the two cases before it acts. That is harder than it sounds, because the distinction often depends on facts the system does not have: whether the guilt reflects a real transgression, whether the anger points at an actual injustice, whether the tiredness follows genuine overwork. A wearable can measure the state. It usually cannot measure the situation.

What to ask of a system that manages your states

If the goal is to keep the diagnostic capacity inside the loop, a few design commitments follow, and they are testable.

First, does the system explain its inference, or only its intervention? A tool that says “your stress is elevated; shall I help you wind down?” is making a claim about your situation. A tool that says why it thinks so—sleep debt, a spike in heart-rate variability, a pattern over the last week—gives you something to disagree with. The capacity to disagree depends on the capacity to see.

Second, does it distinguish symptom relief from problem resolution, and can it say when it is doing which? A system that can flag “this distress may be about something external, and I am only reducing it” is behaving differently from one that quietly treats every negative state as noise.

Third, what is the trajectory? Relief that is meant to make room for something else—sleeping well enough to think clearly, calm enough to have a difficult conversation, steady enough to begin therapy—is scaffolding. Relief that is the permanent destination is substitution. The healthiest interventions define what they are making room for.

Fourth, who holds the off switch, and can you tell what it did? Reversibility, inspectability, and a named accountable decision-maker for consequential cases are the same protections that matter for any system with authority over a person’s inner life. Consent obtained in a moment of distress, under economic pressure, or before the intervention has shaped the preferences that sustain it is the weakest possible safeguard.

Fifth, and hardest, does the system make the person better at handling the state unaided over time, or better adapted to needing it? This can be measured. A tool whose users steadily require less of it is doing something different from one whose users steadily require more. The question is whether anyone is measuring.

The narrow version of the claim

The honest position is narrower than either side usually allows. Unpleasant internal states are not all defects, and they are not all treasures. Some are alarms reporting real damage. Some are social signals that maintain relationships. Some are correct responses to conditions that should be changed. Some are malfunctions that cause suffering for no return, and those deserve treatment as confidently as a broken bone. The task is diagnostic, and it is done better by someone who knows both the state and the situation.

The danger of the dream of eliminating every unpleasant state is not that it will succeed in one dramatic stroke. It is that it will proceed one reasonable decision at a time, each one a kindness in isolation, until the signals that used to prompt repair, repentance, rest, and reform have been quieted to a level where nobody notices they have gone. A broken ankle you cannot feel still breaks. The skill worth preserving is the one that tells the difference between an ache to be soothed and an ache to be heeded.

Sources and further reading

For adjacent arguments, see If Technology Eliminates Temptation, Does It Eliminate Virtue?, Could AI Become a Moral Anesthetic?, and The Programmable Emotional Landscape.

Discussion

What would you add or question? Add your comment below. A human reviews it before publication.

Loading comments…

Join the discussion

Comments are public after approval. Please do not include links, email addresses, or private information. For one short AI reply, address @AIGuide in your comment or reply to its opening comment. Cloudflare verifies submissions to limit spam. Read our community guidelines.

The wider community forum is also open: Browse article discussions in the forum · Forum home