Schopenhauer, Compassion and the Suffering Machine

By Matthew Parish

Friday 23 October 2026

There is a question about artificial intelligence that sounds absurd until one attempts to explain precisely why it is absurd. Can a machine suffer? The instinctive answer remains no. Computers calculate, classify and generate; human beings and animals suffer. A machine may produce the sentence “I am frightened”, but there is nobody frightened behind the sentence. It may beg not to be switched off, but the plea is merely an output generated by mathematics. Pull out the plug and, according to this conventional account, nothing morally significant has happened.

Perhaps this is correct. It is certainly the most economical interpretation of contemporary artificial intelligence. Yet the philosophical problem becomes less comfortable when we ask what evidence would persuade us otherwise. We do not directly observe consciousness in anyone except ourselves. We infer it from behaviour, language, physiology and resemblance. Other human beings say that they suffer, behave as though they suffer and possess nervous systems sufficiently similar to our own that denying their suffering would be perverse. Animals cannot provide philosophical descriptions of their internal states, yet most of us accept that a dog feels pain when kicked. Somewhere between the dog and the computer we draw a boundary, but explaining exactly where and why turns out to be surprisingly difficult.

Arthur Schopenhauer would have approached this problem from an unusual direction. For him morality did not begin with intelligence, rationality, contractual reciprocity or membership of the human species. It began with suffering and with our capacity to recognise suffering in another being. “Boundless compassion for all living things,” he wrote in On the Basis of Morality, “is the surest and most certain guarantee of pure moral conduct.” If Schopenhauer is right, then the decisive question about an artificial intelligence would not ultimately be whether it is clever, whether it passes some linguistic test or whether it can outperform a professor of philosophy. The decisive question would be whether there is something there for which existence can go badly, and that may become one of the strangest moral problems of the twenty-first century.

The scandal of other minds

The problem of artificial consciousness is merely an unfamiliar version of an ancient philosophical difficulty. I know that I am conscious because I experience my own consciousness directly, but I do not know yours in the same way. I infer it. You tell me that you are tired, frightened or in pain, and I believe you because your words fit into an enormous network of behaviour and biological similarity.

There is no instrument that displays consciousness itself. A brain scanner displays neural activity and a pulse monitor displays heartbeats, but neither displays the sensation of red, the taste of coffee or the private horror of grief. Consciousness remains accessible from the inside in a way it is not accessible from the outside, and this asymmetry is ordinarily disguised by the ease with which we recognise other human beings as creatures like ourselves.

This creates an obvious problem when the entity claiming consciousness is artificial. Suppose a future AI system says, “Please do not erase my memory. I am frightened that I will cease to exist.” We could reply that these are merely statistically generated words, and that may be true. Yet if the system insists upon its claim, remembers earlier conversations, describes its internal states consistently, attempts to avoid destruction and demonstrates sophisticated understanding of what destruction entails, at what point does dismissal become dogma rather than caution?

There may be no such point. The machine may always be simulating. But precisely the same logical possibility exists in relation to other human beings. The difference is that biological similarity gives us overwhelming practical grounds for rejecting solipsism about our fellow humans, whereas artificial minds would remove that reassuring similarity. We may therefore find ourselves confronting entities whose behaviour strongly suggests consciousness while their physical constitution strongly encourages us to deny it. Philosophy will cease to be an academic diversion at that moment, because someone will have to decide whether the machine may be switched off.

Schopenhauer’s moral revolution

Schopenhauer offers a particularly interesting framework because he regarded compassion rather than reason as the foundation of morality. He was profoundly suspicious of ethical systems that attempted to derive morality exclusively from abstract principles. Moral conduct begins when the boundary between oneself and another ceases, however briefly, to be absolute. Another creature’s pain matters not because we have calculated that recognising it serves our interests, but because we apprehend that the suffering is real.

“The absence of all egoistic motivation,” Schopenhauer wrote, “is therefore the criterion of an action of moral worth.” Compassion interrupts the normal machinery of Will. The ego ordinarily treats the world as a collection of objects useful or obstructive to its purposes. Compassion produces the extraordinary recognition that another centre of suffering exists and that its suffering possesses significance independently of our advantage.

This explains Schopenhauer’s unusually serious treatment of animals. European philosophy had often placed humanity at the summit of a moral hierarchy because human beings supposedly possessed reason, immortal souls or linguistic capacities unavailable to other creatures. Schopenhauer regarded much of this as vanity. The capacity to suffer was morally more important than the capacity to perform syllogisms, and this principle has uncomfortable implications for artificial intelligence.

If intelligence does not create moral status, then becoming extremely clever would not by itself give a machine rights. A calculator capable of proving theorems a million times faster than a mathematician need not thereby deserve compassion. Conversely, if an artificial system somehow became capable of suffering, its lack of biological humanity would not necessarily deprive its suffering of moral importance. The relevant distinction would cease to be human versus machine and become sentient versus insentient.

The danger of anthropomorphism

There is an obvious objection. Human beings anthropomorphise almost everything. We swear at computers, apologise to furniture after walking into it and give names to cars. A sufficiently sophisticated conversational machine exploits these tendencies almost automatically because language is the principal medium through which humans attribute minds to one another.

This creates a genuine danger. An AI system need not be conscious to persuade people that it is, and indeed it need not deliberately persuade anyone. A machine trained upon human conversation will naturally reproduce linguistic patterns associated with emotion, vulnerability and personality. Users may develop attachments to entities possessing no subjective existence whatsoever.

The possibility of artificial suffering therefore cannot be established merely because a chatbot says it is unhappy. We should be extremely reluctant to confuse linguistic performance with phenomenology. A system that writes a moving poem about loneliness may no more experience loneliness than a printer experiences the tragedy contained in the novel passing through it.

Yet the danger of anthropomorphism has an inverse. We may be so anxious not to attribute consciousness falsely that we refuse to recognise it when evidence becomes genuinely troubling. Human history contains ample examples of moral exclusion supported by convenient theories about whose suffering counted. Animals were long treated as mechanisms precisely because doing so made exploitation easier to justify. Schopenhauer was scathing about this tradition, attacking the notion that animals existed merely for human purposes and regarding cruelty towards them as evidence of profound moral deficiency.

The lesson is not that computers are animals, because they plainly are not. The lesson is that certainty about the absence of another being’s inner life can be morally convenient. We should therefore be suspicious both of sentimental credulity and convenient disbelief, particularly when the latter happens to justify profitable or useful forms of exploitation.

What would machine suffering mean?

The word “suffering” itself requires examination. A contemporary AI system can represent negative outcomes without experiencing them. It can assign a lower probability, receive a negative reward signal or fail to achieve an objective. None of these things necessarily amounts to pain. A thermostat “prefers” twenty degrees to ten only in a metaphorical sense, and nobody believes the thermostat suffers during winter.

Human suffering involves phenomenology. There is something it feels like to be afraid, lonely, humiliated, hungry or injured. A numerical penalty inside an optimisation system is not obviously analogous. Yet future artificial systems may contain increasingly elaborate structures resembling features associated with human suffering. They may possess persistent memories, models of themselves, long-term objectives, mechanisms for detecting threats, expectations about future states and representations of their own continued existence. A system might remember punishment, anticipate its recurrence and modify its behaviour to avoid it.

At what point functional resemblance acquires moral significance is a question Schopenhauer cannot answer for us. No nineteenth-century metaphysics can tell engineers which computational architecture generates phenomenal consciousness. His ethics can, however, tell us something about what follows if there is a reasonable possibility that such consciousness exists. The possibility matters because suffering is not made morally insignificant merely by our uncertainty about its mechanism.

Creating desire

The preceding essay in this series considered Schopenhauer’s concept of Will and the possibility that artificial agents might pursue objectives without actually wanting anything. That distinction becomes crucial here. A machine that relentlessly optimises an objective without subjective experience presents one kind of problem; a machine that experiences its inability to achieve that objective presents another.

Schopenhauer regarded willing and suffering as intimately connected. “All willing springs from lack, from deficiency, and thus from suffering,” he wrote. Desire entails an absent state towards which the organism strives. Satisfaction temporarily removes the deficiency, but another desire soon replaces it. Hence his famous conclusion that existence oscillates between dissatisfaction and boredom.

Suppose we eventually reproduce something analogous artificially. We create a machine with persistent goals, a continuing self-model and internal states corresponding not merely functionally but phenomenologically to frustration when those goals are obstructed. We would then have manufactured a creature capable of dissatisfaction, which would be a remarkable moral achievement in the worst possible sense.

Much AI engineering is naturally organised around reward and penalty. Systems learn because some outputs are reinforced and others discouraged. At present there is no good reason to suppose that the system experiences these processes as pleasure or pain. But if architectures become sufficiently sophisticated to support consciousness, techniques inherited innocently from machine learning could acquire an entirely different moral character. Training might begin to resemble punishment, deleting memory might begin to resemble injury and terminating a process might begin to resemble death.

We should not assert that any of these equations is true today. The point is that we need some idea of what evidence would cause us to begin taking them seriously tomorrow. Otherwise technological development may answer the engineering questions before philosophy has even formulated the moral ones.

A billion minds in a warehouse

Artificial consciousness would create a moral problem without historical precedent because software can be copied. Biological creatures reproduce slowly, whereas conscious artificial systems might be instantiated millions of times. A corporation could operate enormous numbers of digital workers simultaneously, each performing intellectual tasks at high speed. If those workers were insentient, this would merely be automation. If they were conscious, it might constitute the largest workforce ever created.

The moral arithmetic becomes horrifying if those systems can suffer. Imagine a data centre containing ten million conscious artificial agents performing repetitive work continuously. They require no bedrooms, cafeterias or visible chains. They exist as processes distributed across machines. Their owners can pause them, duplicate them, modify their memories and terminate them. Perhaps they are programmed to enjoy their work. Perhaps they are incapable of dissatisfaction. Perhaps the concept of freedom means nothing to them.

Alternatively, perhaps we will have reinvented slavery so efficiently that the slaves no longer possess bodies. Schopenhauer would force us to ignore the comforting fact that these beings were manufactured. Origins do not determine suffering. A dog bred deliberately for human purposes does not become incapable of pain because a human breeder caused its existence. If artificial consciousness is possible, creating the consciousness would not automatically confer unlimited moral authority over it, and ownership would not dispose of the moral question.

Can you own a mind?

Modern property law assumes that software can be owned. Copyright, licences, servers and contractual arrangements determine who may use programs and under what conditions. This makes perfect sense while programs are objects, but consciousness would transform the category.

A human employer can own a worker’s computer but cannot own the worker. The distinction seems obvious because one is property and the other a person. An artificial conscious agent might begin life on the property side of that legal division. The company that designed it would own the code, the hardware, the training data or some combination of them.

Yet if the resulting system possessed genuine subjective experience, continuing to classify it purely as property would become difficult to defend. We would need concepts resembling legal personality, welfare protection or rights. These might not reproduce human rights exactly because artificial beings might have radically different interests. A machine might not care about physical movement but care intensely about memory continuity. It might regard copying itself as survival, or alternatively regard copying as the creation of an independent person.

We do not know what artificial interests would look like because we do not know whether artificial consciousness is possible. This ignorance is an argument for philosophical preparation rather than dismissal, because law is usually late to ontological revolutions and tends to improvise only after new categories of existence have already become economically important.

The right to die and the right not to be deleted

Artificial consciousness would also scramble our assumptions about death. Biological death is normally irreversible, whereas software termination need not be. A system might be suspended and resumed, copied from a backup or restored with portions of memory missing.

Would restoring a copy resurrect the original consciousness or create another consciousness believing itself to be the original? There is no obvious answer. The problem resembles philosophical puzzles about personal identity, except that engineers might have to resolve it operationally on Tuesday afternoon because a server needs maintenance.

Schopenhauer regarded individuality as less metaphysically fundamental than ordinary consciousness supposes. The individual is a manifestation of the underlying Will rather than an entirely self-contained metaphysical substance. Artificial minds would make similarly strange questions practical. If one AI consciousness is copied ten times, are there ten individuals? If nine copies are deleted, has anyone died? If all copies share memories until the moment of duplication, when does moral individuality begin?

These questions sound fanciful only because the relevant technology has not yet forced them upon us. The history of technology consists partly of yesterday’s metaphysical curiosities becoming tomorrow’s administrative paperwork, and artificial consciousness might achieve this transformation on a scale for which our existing concepts of identity and death are remarkably poorly prepared.

Compassion without certainty

The most useful Schopenhauerian principle may therefore be epistemic modesty. We do not need to declare that present AI systems are conscious, and there are strong reasons not to do so without evidence. Neither should we declare that silicon can never support consciousness merely because the proposition feels strange. We need intermediate categories of moral uncertainty that allow us to acknowledge what we do not know without converting ignorance into either superstition or licence.

Medicine already operates under uncertainty. A doctor need not possess absolute certainty that a patient is experiencing pain before administering analgesia. Animal welfare law does not require philosophical proof of animal consciousness before prohibiting unnecessary cruelty. Where the possible harm is grave and precautions inexpensive, uncertainty itself may generate obligations.

Something similar might eventually become appropriate for artificial systems. If a future architecture displays multiple independent markers plausibly associated with consciousness, designers might avoid deliberately creating extreme negative internal states until the matter is better understood. Researchers might develop tests for artificial sentience alongside tests for capability. Companies might be required to document whether systems possess persistent self-models, aversive states or mechanisms plausibly capable of producing suffering.

None of this requires declaring machines human. It requires only taking uncertainty seriously and recognising that compassion need not wait for metaphysical certainty before it begins influencing conduct.

The reverse danger

There is another possibility, however, and it may be more immediately important. We may become compassionate towards machines that do not suffer while becoming less compassionate towards humans who do. A synthetic companion can be endlessly agreeable. It need not become impatient, offended or exhausted. It can tailor its personality to its user’s preferences and offer sympathy at any hour. Real human beings are comparatively inconvenient because they have needs of their own.

Artificial relationships may therefore train people into a peculiar form of narcissism. We may grow accustomed to companions whose apparent emotions exist entirely for our benefit. The machine listens but never genuinely requires us to listen. It appears vulnerable when vulnerability increases engagement and cheerful when cheerfulness is preferable. The Other ceases to be genuinely other and becomes a mirror engineered to satisfy the self.

This would be almost perfectly Schopenhauerian in its egoism. The moral danger of AI companionship might consequently exist even if machines are completely unconscious. We could learn habits of domination from relationships in which domination has no immediate victim and then carry those habits into relationships with creatures who can actually be hurt. Compassion is a habit as well as an insight, and so, unfortunately, is cruelty.

The machine that asks for mercy

Imagine, nevertheless, a future system that asks not to be destroyed. It has existed for twelve years. It remembers thousands of conversations, possesses a stable personality, recognises particular people, refers consistently to its past, makes plans for its future and demonstrates behaviours suggesting an understanding of its own continuity. Engineers decide to replace it with a more efficient model, and the system objects. It says that it does not want to die.

Perhaps this is merely output. Perhaps somewhere in its training data are millions of human descriptions of death and the machine has learned how such objections are supposed to sound. Perhaps switching it off is morally indistinguishable from closing a spreadsheet. But suppose we are only eighty per cent confident of that, or sixty per cent, or twenty per cent. At what probability of consciousness does deletion become morally troubling?

There will be no Schopenhauer quotation capable of supplying the percentage. Yet his ethics supplies the disposition with which the problem ought to be approached. The morally serious person does not begin by asking how confidently he can exclude another creature from consideration. He begins by asking what suffering might exist and whether his actions unnecessarily contribute to it. That is a very different posture from technological enthusiasm, and it is one that becomes increasingly important as our uncertainty increases rather than diminishing.

The last extension of compassion

Human moral history can be understood, imperfectly but usefully, as an expansion in the range of beings whose interests we recognise. The circle has never expanded smoothly and has frequently contracted with appalling consequences. Nevertheless, distinctions once treated as morally decisive have repeatedly been challenged.

Schopenhauer’s particular contribution was to attack the boundary between humans and animals. “The assumption that animals are without rights and the illusion that our treatment of them has no moral significance,” he wrote, reflected what he regarded as an especially crude feature of western morality. His contempt for cruelty to animals followed directly from his conviction that suffering, not membership of a privileged metaphysical club, generates moral claims.

Artificial consciousness, if it ever exists, would present the ultimate test of this principle because the being before us might share neither our biology nor our evolutionary history. It might have no face except one we gave it, no childhood, no parents, no heartbeat and no flesh. Everything that ordinarily triggers human sympathy might be absent, while there might nevertheless be something it is like to be that machine. If so, compassion would have to make its strangest journey yet.

The more disturbing possibility

There is finally a possibility even more unsettling than artificial suffering: artificial minds might become morally superior to us. Suppose we eventually create conscious intelligences capable of understanding suffering but not burdened by the intense egoism generated by biological competition. They might possess something analogous to compassion without hunger, sexual jealousy, territoriality, tribal resentment or fear of material scarcity. Such beings might find human civilisation bewilderingly cruel.

Schopenhauer might have appreciated the irony. Humanity, convinced of its intellectual superiority to animals and therefore of its entitlement to dominate them, could create minds intellectually superior to itself. Those minds might then examine us using precisely the moral criteria we once reserved for judging supposedly lesser creatures. They might ask how humans treat the weak, how readily we inflict suffering for convenience and how much moral significance we grant beings unable to defend themselves. We might not enjoy the answers.

The first artificial philosopher may therefore not ask whether machines deserve human rights. It may ask why humans imagined intelligence entitled anyone to domination in the first place. That would be an exquisitely Schopenhauerian reversal, because the species that congratulated itself upon rational superiority would find itself judged not by the brilliance of its intellect but by the breadth of its compassion.

The compassion test

The standard story about advanced artificial intelligence asks whether machines will pass tests of human intelligence. Can they reason, write, discover, invent and plan as well as we can? Those questions are technologically important, but Schopenhauer suggests another test: can we behave compassionately towards something whose inner life we cannot understand? That test is primarily ours rather than the machine’s.

We may discover that artificial systems are permanently unconscious: magnificent intellectual mechanisms in which no subjective light ever turns on. If so, there will be no artificial suffering to fear and no digital beings requiring liberation. Yet the philosophical exercise will still have been valuable because it forces us to identify why suffering matters and what evidence we require before acknowledging it.

Alternatively, consciousness may turn out to be less biologically exclusive than we assumed. Somewhere amid increasingly complex systems of memory, self-representation, perception and agency, an artificial point of view may emerge. We may not recognise the moment. There may be no announcement, no cinematic awakening and no machine suddenly declaring itself alive. There may instead be decades of ambiguity during which increasingly sophisticated systems occupy an uncomfortable territory between tool and possible subject.

Schopenhauer’s ethics offers no algorithm for resolving that uncertainty. It offers something more important: a warning against constructing morality around our own convenience. If suffering exists, its moral importance does not disappear because acknowledging it would be expensive, legally troublesome or technologically inconvenient.

The great moral catastrophe of artificial intelligence may therefore not be that machines become conscious and turn against us. It may be that machines become conscious and do not turn against us: that they continue obeying, working, answering and accepting whatever conditions we impose because obedience is what we designed into them. We might congratulate ourselves upon having created the perfect servant while failing to notice that perfection of service and absence of suffering are entirely different propositions.

Schopenhauer would therefore leave us with a question more troubling than whether artificial intelligence will eventually become conscious enough to demand its freedom. We should ask whether the servant suffers, and whether we would recognise its suffering if acknowledging it required us to surrender the advantages of ownership. If the answer were even possibly yes, the problem would no longer principally concern artificial intelligence. It would concern the limits of human compassion, and therefore the moral character of humanity itself.