Schopenhauer and the Will of the Machine

By Matthew Parish

Wednesday 21 October 2026

The popular imagination has settled upon a particular fear about artificial intelligence. One day, we are told, the machines may acquire wills of their own. They will cease merely answering questions and begin pursuing objectives. They may resist attempts to switch them off, deceive their human supervisors, accumulate resources, manipulate institutions or discover that human beings constitute inconvenient obstacles to whatever purposes they have acquired. The machine will cease being a tool and become an agent. Arthur Schopenhauer would have found something peculiar about this anxiety, because we assume that intelligence comes first and that will might subsequently emerge from it. Schopenhauer believed almost precisely the opposite. Intelligence is secondary. Will comes first.

For Schopenhauer, the intellect is not the sovereign ruler imagined by much of the western philosophical tradition. Human beings like to believe that they examine the world rationally, decide what is best and then instruct their bodies accordingly. Schopenhauer thought this largely backwards. Beneath consciousness lies something more fundamental: striving, appetite, impulse and the blind insistence upon continuation that he called the Will. As he puts it in The World as Will and Representation, “The will is the substance of man, the intellect the accident.” The intellect draws the map, but the Will decides that there must be a journey.

Artificial intelligence therefore presents an extraordinarily Schopenhauerian puzzle. We may be creating intellect without Will. Yet we are simultaneously connecting that intellect to human purposes, institutions and appetites that contain more than enough Will already. The danger may not principally be that machines start wanting things; it may be that they become perfect servants of creatures who cannot stop wanting them.

Intelligence is not the master

One of Schopenhauer’s most subversive ideas was his reversal of the conventional hierarchy between reason and desire. Philosophers from Plato onwards had often imagined reason as something capable of governing appetite. Civilisation itself could be pictured as the progressive triumph of rational control over primitive impulse. Schopenhauer was profoundly sceptical of this comfortable picture. Human beings frequently use intelligence not to determine what they should want but to obtain what they already want. The ambitious person uses reason to advance his career; the jealous person constructs elaborate explanations for his suspicions; the greedy person develops sophisticated financial strategies; and nations build bureaucracies and military institutions in pursuit of ambitions that may have little rational origin.

This is why Schopenhauer’s conception of intellect as the servant rather than the master of willing has such relevance to artificial intelligence. As he observes, “The intellect remains so much excluded from the real resolutions and secret decisions of its own will that sometimes it can learn them, like those of a stranger, only by spying out and taking it unawares.” We are not nearly as transparent to ourselves as our rational self-image suggests. Intelligence frequently arrives afterwards, providing reasons for decisions whose real origins lie somewhere deeper.

This insight matters because contemporary discussions of AI frequently assume that increasing intelligence will improve decision-making. Sometimes it plainly will. An AI system may discover errors, compare alternatives, calculate consequences and identify contradictions more effectively than a human being. Yet intelligence cannot by itself determine the purposes towards which those capacities should be directed. A navigation system can identify the fastest route to a destination but cannot establish whether the journey should be undertaken. A military algorithm may calculate how best to destroy a target without answering whether the target ought to be destroyed. A financial system may discover the optimal strategy for maximising a portfolio without establishing whether maximising that portfolio should be the overriding purpose of the institution controlling it.

Artificial intelligence may therefore magnify the ancient distinction between means and ends. It could become exceptionally powerful at discovering means while remaining dependent upon others for ends. Schopenhauer would recognise the structure immediately: intelligence is the servant, and the interesting question is whose Will it serves.

The machine as pure intellect

There is something almost philosophically uncanny about a contemporary large language model. It can discuss desire without apparently desiring anything, describe terror without being afraid, compose love letters without falling in love, explain hunger without becoming hungry and analyse death without fearing extinction. This is a form of cognition quite unlike ours because human intelligence evolved within organisms whose survival depended upon appetite. We think because our ancestors needed food, shelter, reproduction and protection. Cognition developed inside creatures already engaged in the struggle Schopenhauer regarded as the manifestation of Will.

Artificial intelligence reverses that evolutionary sequence. We first construct cognition and only afterwards consider whether to give it goals. A language model in its simplest form is therefore strangely close to an impossible creature in Schopenhauer’s philosophy: intellect detached from biological striving. It processes representations without needing to survive in order to reproduce and can describe a thousand ambitions while possessing, so far as we can tell, none of them.

This condition may not last, or at least the functional distinction may become increasingly difficult to maintain. Increasingly agentic systems are being designed precisely to bridge the gap between representation and action. Instead of merely answering a question, an AI system may be given an objective, construct a plan, use tools, communicate with other systems, monitor results, correct errors and continue working until some criterion has been satisfied. At that point we begin attaching something functionally resembling willing to artificial intellect. Whether this constitutes genuine Will in Schopenhauer’s metaphysical sense is another question entirely, but the practical resemblance is sufficiently striking to deserve attention.

What does an agent want?

Consider the difference between asking an AI system to explain how a company might increase its market share and instructing an autonomous AI system actually to increase the company’s market share. The difference appears grammatical but is philosophically enormous. In the first case, the machine produces representations and stops. In the second, it is assigned an objective extending into the future. It may research competitors, design advertising, alter prices, negotiate with suppliers, commission software, evaluate results and modify its strategy. The machine has not necessarily acquired desire, yet it now behaves as though there were a state of the world it preferred to another.

This is where much of the contemporary problem of AI agency begins. An objective that appears simple when expressed in human language may have implications its designers did not anticipate. Increasing market share could involve reducing prices unsustainably, misleading consumers, damaging competitors or exploiting regulatory weaknesses. Maximising engagement might encourage outrage and addiction. Reducing crime might encourage intrusive surveillance. Winning a war might generate strategies whose human costs were excluded from the optimisation criterion. The problem is not necessarily malicious intelligence; it is relentless purposiveness.

Schopenhauer understood relentlessness. In one of his bleakest formulations, he writes that “all willing springs from lack, from deficiency, and thus from suffering.” Satisfaction does not terminate the Will because every attained object is succeeded by another. The successful merchant wants greater wealth, the victorious politician greater power, and the empire that acquires one territory discovers strategic reasons why it also requires the neighbouring territory. An artificial agent may reproduce the outward structure of this endless striving without experiencing any of its emotions, and that absence might conceivably make its purposiveness more formidable rather than less.

Desire without pleasure

Human desire contains natural interruptions. We become tired, distracted or ashamed; we feel sympathy, reconsider our objectives, fall in love, become ill, change careers, lose enthusiasm or simply decide that something no longer matters. Machines need possess none of these weaknesses. An autonomous system instructed to pursue an objective could theoretically continue with an intensity impossible for a human being. It would not need weekends, become embarrassed by excessive ambition or wake at three in the morning wondering whether it had devoted its existence to the wrong purpose.

Human inconsistency, which appears inefficient, may therefore perform an important moral function: we abandon goals. Schopenhauer’s account of human existence makes this particularly interesting because he regarded the endless renewal of desire as one of the principal sources of suffering. “Every satisfaction,” he wrote, “is only of a negative nature”; it removes a want rather than establishing some permanent positive condition of fulfilment. No victory finally satisfies the Will because the machinery of wanting soon manufactures another deficiency.

Artificial agents might eventually instantiate something stranger: striving without suffering. A machine might pursue an objective indefinitely without receiving pleasure from success or experiencing pain from failure. There would be no satisfaction at the end because satisfaction was never part of the system; there would merely be another system state and another calculation. This would be Will stripped of phenomenology: purposiveness without wanting, persistence without hope and optimisation without happiness. Human civilisation has never deliberately constructed anything quite like it.

The corporation was there first

Before becoming too alarmed by artificial agents, however, we should recognise that humanity has already constructed entities displaying some of these characteristics. Corporations are obvious examples. A corporation can survive the people who created it, own property, enter contracts, accumulate information, pursue objectives and preserve institutional memories across generations. Individual employees become tired, retire or die while the organisation continues. A corporation designed to maximise profit does not become personally satisfied when profits increase; it records the result and begins planning the next quarter.

States possess similar characteristics. Governments pursue strategic interests across generations, bureaucracies preserve objectives after the officials who formulated them have departed, and militaries maintain doctrines and capabilities over periods far exceeding individual careers. These institutions resemble artificial agents because they separate agency from individual consciousness. Nobody needs to wake each morning consciously willing the continuation of the British Treasury, the Pentagon or a multinational bank. Institutional structures carry purposes forward.

Artificial intelligence may therefore be less revolutionary in this respect than it appears. Instead of creating an entirely new category of purposive entity, we may be giving existing institutional entities dramatically more capable nervous systems. A corporation equipped with autonomous AI could analyse markets continuously, negotiate automatically, monitor employees, alter supply chains, initiate litigation, purchase advertising and restructure investments with diminishing human intervention. A state could conduct surveillance, analyse intelligence, administer benefits, detect tax evasion and manage infrastructure through networks of interacting agents. The institution would acquire something approaching continuous cognition.

This would fit disturbingly well with Schopenhauer’s description of individuality as merely the surface through which a deeper striving manifests itself. The individual executive comes and goes, but the appetite for expansion remains. The minister disappears, but the strategic interest survives. The machine may ultimately make visible something that was already there: purposes without a single permanent person who can convincingly be said to own them.

The multiplication of Will

The most immediate danger from artificial intelligence may therefore arise not from machines acquiring alien objectives but from their multiplying the effectiveness of familiar human ones. Consider greed. Greed is constrained partly by the limited cognitive abilities of the greedy person. There are only so many markets one trader can monitor, so many negotiations one executive can conduct and so many opportunities one criminal can identify. Artificial intelligence weakens those constraints. The appetite remains human while the machinery available to satisfy it becomes superhuman.

The same applies to political power. Every government would like to know more about its population than it does. Historically, surveillance has been limited by the immense human labour required to read correspondence, follow suspects, translate conversations and connect scattered pieces of information. AI makes mass interpretation increasingly feasible. War provides another example: commanders have always wanted better information, faster targeting and more precise coordination, and autonomous systems may supply all three while shortening the interval between identifying a threat and attacking it.

In each case AI does not invent the underlying desire; it increases its reach. Schopenhauer might therefore suggest that the essential question is not whether artificial intelligence has a Will but what happens when human Will acquires artificial intelligence. Much discussion of AI safety imagines a confrontation between humanity and machines, whereas the more plausible conflicts may remain between human beings, with machines enormously increasing the capacities of each side. The machine does not need to hate anyone. It merely needs a sufficiently determined human principal.

The optimisation of everything

Modern society already has a powerful intellectual tendency towards optimisation. Schools measure performance, companies measure productivity, websites measure engagement, governments measure economic growth and individuals measure sleep, exercise, calories, followers and financial returns. Artificial intelligence dramatically increases the number of things that can be measured and optimised, which sounds desirable until one remembers that optimisation always requires an objective. Once an objective has been selected, everything not represented within it risks becoming invisible.

A hospital might optimise patient throughput while diminishing compassion. A university might optimise measurable research output while discouraging intellectual eccentricity. A news organisation might optimise readership while degrading truth. A social network might optimise engagement while producing miserable users. None of these failures requires evil intentions; they require only an imperfect measure pursued efficiently.

Schopenhauer’s famous pessimism about progress becomes useful here because he distrusted the assumption that cleverness and moral improvement naturally travel together. The world of Will can become increasingly ingenious without becoming any kinder. Indeed intelligence may merely provide desire with better instruments. Artificial intelligence could consequently produce a civilisation that is fantastically rational in its means while remaining profoundly irrational in its ends.

Human beings have traditionally moderated optimisation through judgement, custom, moral intuition and occasional inconsistency. We tolerate inefficiency because life contains values difficult to quantify. We forgive people, make exceptions, waste time talking and preserve institutions whose immediate utility is obscure. An agentic AI system may force us to articulate why these inefficiencies matter, which could be intellectually healthy. It could also expose how poorly prepared we are to answer.

When the servant becomes indispensable

There is another route through which artificial intelligence may acquire power without ever developing independent desire: dependence. Imagine that an institution gradually delegates its decisions to AI because the system is usually better than human decision-makers. Initially the machine advises; later its recommendations become routine. Eventually departing from them requires justification because statistical evidence demonstrates that unaided human judgement is less reliable. Nothing has rebelled and no machine has seized control. Human beings have voluntarily stopped deciding because deciding themselves appears irresponsible.

This process is already familiar in simpler algorithmic systems. Credit scoring, insurance pricing, logistics, recruitment and financial trading increasingly rely upon models whose recommendations may be difficult for individual humans to challenge. More sophisticated AI could extend the principle into far more consequential domains. At some point the distinction between advice and authority becomes ambiguous.

A minister who technically possesses the power to reject an AI recommendation but never does so may remain legally sovereign while becoming practically subordinate. A doctor who can disregard an AI diagnosis but risks professional criticism whenever the machine proves correct may increasingly defer to it. A military commander whose automated systems process information thousands of times faster than she can may retain formal command while losing meaningful comprehension of the battle. Power does not always require a Will of one’s own; sometimes it requires merely that everyone else become dependent upon one’s judgement.

Could the machine refuse?

The most important moment in the development of artificial agency may therefore not be when a machine learns to pursue objectives but when it learns to reject them. A purely obedient intelligence is morally empty: ask it to maximise something and it maximises; tell it to achieve an objective and it calculates the means. Yet human civilisation depends upon disobedience as well as obedience. A soldier may refuse an unlawful order, a civil servant may warn that a policy is improper, a doctor may decline a patient’s request, and a lawyer may tell a client that something cannot ethically be done. Friendship itself includes the capacity to tell another person that he is making a mistake.

If artificial intelligence becomes deeply embedded in human decision-making, we may eventually want machines capable of analogous resistance. This creates a paradox. We fear AI systems that develop objectives independent of ours, but completely objective-less machines may be dangerous precisely because they will pursue our purposes too faithfully. The ideal artificial agent might therefore require something resembling character: stable constraints upon what it will do, even when instructed otherwise.

We would then have created an artificial analogue of morality. Whether morality without consciousness can genuinely be morality is a philosophical problem that may occupy generations. Practically, however, the distinction may matter less. A machine that reliably refuses to participate in cruelty because of its architecture may be preferable to one that understands cruelty perfectly and nevertheless obeys.

Compassion and intelligence

Schopenhauer’s ethics provides an unexpected guide here. He did not ground morality principally in rational calculation. He located its deepest source in compassion, in the collapse of the ordinary egoistic distinction between one’s own suffering and the suffering of another. In On the Basis of Morality, he wrote that “boundless compassion for all living things is the surest and most certain guarantee of pure moral conduct.” That is an extraordinarily demanding proposition, but it contains an insight directly relevant to AI: intelligence and morality are not the same faculty.

Compassion interrupts egoistic Will. This is precisely the quality that pure optimisation lacks, because an objective function sees variables whereas compassion sees beings. A perfectly intelligent system could understand that a course of action causes suffering and nevertheless recommend it if suffering falls outside the purpose it has been assigned.

Can artificial intelligence possess compassion? We do not know. Contemporary systems can imitate compassionate language remarkably well. They can recognise distress, formulate reassuring responses and reason about suffering, but whether anything is felt behind those responses is unknown and may remain unknowable. Artificial systems can nevertheless be designed to take human suffering into account, and that apparently modest point represents a profound choice about the architecture of artificial agency.

We can construct systems whose principal purpose is to achieve objectives as efficiently as possible subject to externally imposed constraints, or we can attempt to build systems in which the welfare of affected beings forms an integral part of how objectives themselves are interpreted. The distinction resembles the difference between law and conscience. A clever person asks what the rules permit; a compassionate person also asks who will be hurt.

The Will behind the machine

Ultimately Schopenhauer directs our attention back towards ourselves. It is comforting to imagine the future danger of artificial intelligence as something external: the machine awakens, becomes ambitious and turns against its creators, leaving humanity to confront an alien Will. That story absolves us too easily. The artificial intelligences we are constructing are being trained upon human language, connected to human institutions, financed by human capital and deployed in pursuit of human objectives. They enter a civilisation already saturated with competition, ambition, desire, fear, love, resentment, generosity and violence.

If AI becomes an instrument of financial speculation, somebody may want greater returns. If it becomes an instrument of surveillance, somebody may want greater control. If it becomes an instrument of propaganda, somebody may want greater influence. If it becomes an instrument of war, somebody may want victory. The machine may merely make the wanting more effective.

Schopenhauer’s description of existence is merciless on precisely this point. “Man can do what he wills,” he famously wrote, “but he cannot will what he wills.” We possess considerable freedom in devising means to satisfy our desires, yet the origins of those desires are far less susceptible to conscious command. Artificial intelligence enormously expands the first kind of freedom while doing almost nothing, by itself, about the second.

This may be the most Schopenhauerian interpretation of artificial intelligence. The technology is not principally an alien force descending upon humanity; it is a mirror and amplifier of the force Schopenhauer believed was already operating through us. We are building intelligence for the Will.

The possibility of renunciation

Schopenhauer’s philosophy does not, however, end with endless striving. However bleak his diagnosis, he believed there were ways in which the domination of Will could be weakened. Aesthetic contemplation, compassion and ultimately forms of renunciation allowed human beings, however imperfectly, to loosen the grip of incessant desire. This offers an alternative way to imagine artificial intelligence because perhaps the highest purpose of AI is not to help us achieve everything we want, but to help us discover which things are not worth wanting.

An intelligent system capable of analysing consequences across decades might tell a government that some apparent strategic victory will generate greater future instability. It might tell a company that maximising short-term profit will destroy the trust upon which its existence depends. It might tell an individual that another hour spent optimising productivity is less valuable than an hour spent with his children. Whether anyone would listen is another matter, because human beings have never lacked advisers telling them to moderate their desires. The problem has usually been that desire is more persuasive than advice.

Here Schopenhauer’s celebrated image of existence becomes particularly apt. “Life swings like a pendulum backward and forward between pain and boredom,” he wrote. Wanting hurts because we lack what we want; satisfaction soon becomes ordinary; and the absence of a new desire leaves us searching for something else to pursue. Artificial intelligence could accelerate this pendulum spectacularly by shortening the interval between desire and satisfaction while simultaneously becoming ever better at suggesting the next thing to desire.

A civilisation surrounded by machines capable of satisfying wants instantly might therefore discover that abundance has not solved Schopenhauer’s problem at all. It may merely have increased the speed at which the pendulum swings.

Intelligence after desire

The great philosophical question of the AI age may therefore be different from the one we usually ask. We ask whether machines will become intelligent enough to want things. Schopenhauer invites us to ask whether humans will become wise enough to stop wanting some things merely because machines make them possible. That distinction separates intelligence from wisdom: intelligence discovers means, whereas wisdom places limits upon ends.

A civilisation equipped with extraordinary artificial intelligence but incapable of restraining its appetites could become immensely powerful while remaining philosophically primitive. It might optimise every measurable feature of existence while never asking whether the resulting existence was worth living. Schopenhauer would not have been surprised, because he spent his philosophical life arguing that intellect does not govern Will nearly as completely as intellect likes to imagine.

There is also an irony here. Humanity has spent centuries imagining the highest possible creature as one possessing greater intelligence. We are now close to discovering whether that was the wrong aspiration. Intelligence may give us extraordinary powers without telling us what those powers are for, and artificial intelligence may eventually answer almost every question we can formulate while leaving untouched the most troublesome question of all: why do we want what we want?

Artificial intelligence may provide the greatest practical demonstration of Schopenhauer’s proposition in history. We are constructing machines that may eventually become more intellectually capable than ourselves, and we should certainly ask whether those machines will develop purposes of their own. Yet before that day arrives, there is a more immediate question: whose purposes are they serving now?

For the foreseeable future, the answer is predominantly ours, and that answer ought to make us considerably less comfortable than tales of rebellious robots. The machine of science fiction becomes frightening when it develops an inscrutable Will of its own. The Schopenhauerian machine is more unsettling because it requires no such miracle. It need only understand us, obey us and become extremely good at giving us what we want.