After AGI: What Are Human Beings For?

By Matthew Parish

Monday 7 September 2026

For most of human history, intelligence has been a scarce commodity. This rather obvious fact has shaped almost every institution we have created. We pay doctors because it is difficult to understand medicine, lawyers because the law is complicated, engineers because bridges ought not to fall down and professors because somebody has to understand Hegel. Governments employ civil servants to absorb information and make decisions. Companies employ managers to organise other people. Universities select the intellectually able and spend years making them more able still. Entire systems of social status have grown around the possession, or purported possession, of unusual powers of reasoning.

Artificial intelligence presents us with the possibility that intelligence will cease to be scarce.

This is a considerably more profound prospect than robots taking people’s jobs. Humanity has experienced technological unemployment before. Agricultural machinery displaced farm labourers, industrial machinery displaced artisans and computers eliminated entire categories of clerical work. Yet each of these revolutions left intact a reassuring assumption: somewhere behind the machine stood a human mind. Machines supplied strength, speed, precision and memory but people supplied judgment.

Artificial general intelligence threatens to overturn that division of labour. If machines can eventually reason better than we can, acquire knowledge faster than we can, remember more than we can and communicate with comparable subtlety, then we confront a question for which neither economics nor philosophy has prepared us particularly well.

What, exactly, are human beings for?

The wrong question

The question is deliberately provocative because human beings are not for anything. A screwdriver is for turning screws and a computer is for processing information because these are artefacts designed for purposes. Human beings have no equivalent manufacturer’s specification.

Nevertheless modern civilisation has quietly taught us to understand ourselves in precisely these instrumental terms. When strangers meet, one of their first questions is invariably “What do you do?” The expected answer is an occupation. We do not normally reply that we raise children, contemplate mortality, enjoy Bach, love somebody or watch the rain. We identify ourselves by our contribution to the economic division of labour.

This habit becomes particularly obvious when people discuss artificial intelligence. The most common anxiety is not that machines will become intelligent but that they will take our jobs. The implication is revealing. We have allowed the concepts of occupation, economic usefulness and personal worth to become entangled. That may prove psychologically disastrous if artificial intelligence develops as rapidly as its advocates expect.

The end of intellectual scarcity

Consider what happens when intellectual labour becomes extremely cheap. A competent lawyer presently represents the accumulated product of decades of education and experience. So does a surgeon, architect, accountant or engineer. Their services are expensive because expertise is difficult to acquire and human time cannot be reproduced.

Software has different economics. Once an intelligent system can perform a particular cognitive task, another copy of that capability can in principle be deployed at negligible marginal cost. Ten excellent human lawyers require ten salaries, ten offices and perhaps a collective century of education. Ten thousand instances of an excellent artificial legal intelligence principally require computing infrastructure.

This does not mean that all lawyers will disappear next Tuesday. Professional regulation, liability, client preferences, institutional inertia and the imperfections of contemporary AI will preserve human employment for a substantial period. Similar qualifications apply to doctors, academics, programmers and accountants. Yet these considerations concern the speed of economic adjustment rather than its ultimate direction.

If artificial intelligence eventually becomes better than humans at most intellectual work, the fundamental economic resource traditionally possessed by educated people — their minds — undergoes radical depreciation.

There is an uncomfortable historical parallel. Before industrialisation, physical strength had considerable economic value. Machinery reduced its importance. A modern crane possesses physical abilities no human being can remotely approach, yet nobody suffers an existential crisis because he cannot lift a shipping container. We may eventually regard intellectual superiority in much the same way.

The humiliation of intelligence

This may be difficult because intelligence occupies an unusually exalted place in our conception of ourselves. We already accept that machines are stronger than us, faster than us and more precise than us. These facts do not threaten human dignity because we have never seriously imagined that our value consists in being able to outrun a motorcar.

Intelligence is different. Aristotle described man as the rational animal. The Enlightenment placed reason at the centre of human emancipation. Liberal political philosophy assumes individuals capable of deliberation and choice. Universities organise elaborate hierarchies around intellectual accomplishment. Even ordinary social life assigns prestige to people perceived as clever.

For centuries we have comforted ourselves with the thought that however physically insignificant Homo sapiens may be, ours is the species that understands the universe. What happens if we are no longer the creatures that understand it best?

The first reaction will probably be denial. We can already observe versions of it. Whenever artificial intelligence masters another intellectual activity, people explain that the accomplishment does not represent real intelligence. A computer defeats the world chess champion: chess was only calculation. A machine writes poetry: poetry requires lived experience. It solves mathematical problems: mathematics is formal manipulation. It diagnoses diseases: diagnosis is pattern recognition.

This process can continue almost indefinitely because the definition of “real intelligence” can always be relocated to whichever territory machines have not yet conquered. Eventually this becomes less an argument about machines than an exercise in protecting human self-esteem.

The centaur period

Before imagining a world in which human intellectual labour has become redundant, we should recognise that there will probably be a lengthy intermediate stage. Human beings and artificial intelligence will work together, each compensating for the other’s weaknesses. This arrangement is sometimes described as the “centaur” model after the hybrid creatures of Greek mythology.

We are already entering this period. A lawyer assisted by advanced AI may analyse authorities faster than an unaided lawyer. A scientist may use AI to explore hypotheses she could not investigate alone. A programmer may supervise machines that write most of the code. A doctor may combine personal knowledge of a patient with diagnostic systems trained upon quantities of medical information no physician could possibly read.

During this stage the relevant competition will often not be human against machine but human-with-machine against human-without-machine. Refusing AI assistance may eventually resemble a mathematician refusing a calculator on the ground that arithmetic ought to be performed authentically.

The centaur period may be extraordinarily productive. It may also be brief. If the artificial component continues improving, one must ask what precisely the human component contributes. Initially the answer may be judgment, contextual understanding, accountability and interpersonal trust. Later the answer may contract to accountability and trust. Eventually even those distinctions may become uncertain.

At some point the centaur discovers that the horse can manage perfectly well without the rider.

Human judgment

One popular answer is that human beings will retain responsibility for decisions because machines cannot possess genuine judgment. There is something important in this argument but it should not be exaggerated.

Judgment is often simply the name we give to reasoning under conditions of uncertainty. A senior diplomat possesses “judgment” because she has accumulated decades of experience about personalities, institutions, incentives and unintended consequences. An experienced barrister possesses it because he has seen how courts actually behave rather than merely reading what the law says.

There is no obvious reason why machines must remain permanently incapable of analogous abilities. If judgment emerges from experience, pattern recognition, probabilistic reasoning and the ability to compare unfamiliar circumstances with previous ones, these appear precisely the sorts of faculties artificial intelligence may acquire.

The more compelling distinction is responsibility. We may decide that certain decisions ought to be made by human beings even where machines could make them better. A judge, general, doctor or elected politician does not merely calculate an answer. He occupies a social role carrying moral and legal responsibility.

That distinction is crucial. Yet it means that the remaining human function may be not superior intelligence but legitimate authority.

Politics after superior intelligence

This raises peculiar political questions. Imagine an artificial intelligence capable of designing a substantially better national budget than the Chancellor of the Exchequer, predicting the consequences of legislation more accurately than Parliament and negotiating trade agreements more effectively than diplomats. Should Britain follow its recommendations?

The temptation would be enormous. Governments are already dependent upon technical expertise. If artificial systems become dramatically better at modelling complex social and economic consequences, politicians who ignore them may begin to look irresponsible.

Yet democracy is not a mechanism for discovering technically optimal answers. It is a system for distributing political authority among human beings. Citizens are entitled to make collective mistakes because the state belongs to them.

The same principle applies internationally. Ukraine may use increasingly sophisticated artificial intelligence to allocate resources, analyse Russian military activity and optimise reconstruction. Yet decisions about what sacrifices Ukraine should make, what risks she should accept and what sort of country she wishes to become cannot legitimately be outsourced merely because a machine calculates consequences more accurately. NATO may employ AI throughout its command structures and the EU may use it in economic administration, but neither organisation acquires moral personality merely because its algorithms become extraordinarily intelligent. Europe likewise remains a human political civilisation rather than an optimisation problem.

This distinction between intelligence and sovereignty may become one of the central political questions of the twenty-first century.

War and the human purpose

War provides the starkest illustration. Modern warfare is already moving towards machines that identify targets, analyse imagery, navigate without continuous human control and coordinate battlefield information at speeds people cannot match.

Ukraine’s experience has accelerated this transformation. She has been compelled by necessity to innovate with drones, electronic warfare, artificial intelligence and distributed battlefield technologies at extraordinary speed. Russia has done likewise. The battlefield increasingly contains machines hunting machines, while human beings become supervisors of systems whose immediate decisions may occur too quickly for meaningful intervention. There is an obvious military logic to this. If a machine can recognise an incoming drone and destroy it in milliseconds, requiring a human being to approve the interception may simply ensure that the target explodes first.

But there is a threshold beyond which automation creates profound moral discomfort. Killing is not merely an engineering operation. A decision to take human life carries moral significance precisely because somebody is responsible for it.

A perfectly accurate autonomous weapon might conceivably kill fewer civilians than frightened soldiers operating under battlefield conditions. Yet we might still hesitate before delegating entirely to it, because efficiency is not the only value at stake. Some things matter because humans choose them.

Love is not a productivity contest

This brings us towards the beginning of an answer. Much of what human beings value most has never depended upon being the best at anything.

A mother does not love her daughter because the daughter represents an efficient allocation of emotional resources. Friendship is not valuable because friends provide optimised conversational output. We do not listen to music because doing so increases gross domestic product.

Human relationships derive their significance from particularity. If somebody loves you, it matters that that person loves you. A machine capable of producing more eloquent expressions of affection does not thereby render human affection obsolete. This principle may extend surprisingly far. Suppose an artificial intelligence could write a better novel than Tolstoy. Would War and Peace cease to matter? Surely not. Part of the significance of art lies in the fact that another human consciousness attempted to communicate something about existence.

A machine may eventually compose a technically superior symphony to Beethoven’s Ninth. Yet Beethoven’s deafness, frustration, mortality and determination are part of why the Ninth means what it does. Human imperfection is not always a defect awaiting technological correction.

The return of Aristotle

Curiously, artificial intelligence may force modern societies to rediscover an ancient distinction between work and life. The Greeks did not generally regard economically productive labour as the highest human activity. Aristotle thought the good life involved contemplation, friendship, political participation and the cultivation of virtue. The modern industrial world inverted much of this hierarchy. Employment became central to identity, while leisure was relegated to the period in which one recovered sufficiently to return to work.

If machines eventually produce most goods and services, we may have to reverse the inversion. Education would then require profound reconsideration. Schools and universities presently prepare people substantially for employment. Students acquire qualifications because qualifications grant access to occupations. If artificial intelligence can perform many of those occupations better than graduates can, education must rediscover purposes beyond labour-market certification.

One possibility is that education becomes more genuinely humanistic. We might study philosophy because understanding how to live is valuable, history because memory matters, literature because other lives enlarge our own and science because the universe is fascinating. That would hardly represent the death of education. It might represent its liberation.

The economic problem remains

There is nevertheless a danger in romanticising this future. A world in which machines perform most economically valuable work does not automatically become a paradise of philosophy, art and leisurely lunches. Someone owns the machines.

If artificial intelligence and robotics generate enormous wealth while ownership remains concentrated among a small number of corporations and individuals, technological abundance could coexist with extraordinary inequality. The majority of human beings might discover not that they have been liberated from labour but that the market no longer has much use for them.

This distinction is fundamental. The social consequences of AGI will depend at least as much upon institutions of ownership, taxation and distribution as upon the underlying technology. Universal basic income is frequently proposed as an answer. Some form of social dividend may indeed become necessary if labour income ceases to be the principal mechanism by which purchasing power is distributed. Governments might tax artificial production, hold sovereign stakes in AI infrastructure or devise entirely new systems of property rights.

Yet distributing money solves only the economic problem. It does not solve the psychological one. A society in which millions of people receive adequate incomes but believe themselves useless may be neither stable nor happy. Human beings need status, community, projects, obligations and reasons to get out of bed. Work presently supplies these things imperfectly but extensively. The post-AGI problem is therefore not merely how to distribute wealth. It is how to distribute purpose.

The danger of artificial purpose

Here lies a darker possibility. If machines become extraordinarily good at satisfying human preferences, they may also become extraordinarily good at supplying synthetic meaning.

An AI companion could always be interested in your stories. Artificial entertainment could be perfectly tailored to your psychology. Virtual environments might supply endless achievement, admiration, romance and adventure without requiring the inconvenience of reality. For some people this might be benign. For others it could become a sophisticated form of sedation.

A civilisation liberated from necessary labour faces two possibilities. People might use their freedom for relationships, exploration, scholarship, creation, public service and contemplation. Or they might disappear into personalised entertainment systems designed to maximise engagement. Aldous Huxley would recognise the danger immediately. The nightmare need not be machines oppressing humanity. It may be machines becoming so extraordinarily good at pleasing us that we cease wanting anything difficult.

Scarcity after abundance

Even in a world of artificial superintelligence, however, certain things will remain scarce.

Time will remain scarce because human lives are finite. Attention will remain scarce because consciousness cannot experience everything simultaneously. Land will remain scarce. Original historical objects will remain scarce. Certain forms of human recognition will remain scarce because there are only so many people whose admiration one can meaningfully possess.

Above all, authentic human relationships will remain scarce because people themselves remain finite. These forms of scarcity may become more important as intellectual and material scarcity decline. Status may migrate away from possession of information towards qualities machines cannot simply mass-produce: courage, loyalty, character, beauty, physical accomplishment, leadership, eccentricity, authenticity and perhaps wisdom. Whether this produces a better society is another matter. Humanity has never lacked ingenuity in devising unpleasant status competitions.

Intelligence was never the point

There is a deeper philosophical mistake in assuming that superior intelligence entails superior value. Albert Einstein was more intelligent than most infants. It does not follow that his life possessed greater moral worth than theirs. Intelligence affects what a being can do. It does not straightforwardly determine what that being is worth.

If machines become vastly more intelligent than human beings, the same distinction applies. Their intellectual superiority would no more establish human worthlessness than a computer’s superiority at arithmetic establishes the worthlessness of a child learning multiplication.

This sounds obvious when stated abstractly. Living through the transition may feel rather different. Entire professional classes have built their identities around being clever. Academics, lawyers, doctors, programmers, financiers and writers may find it particularly difficult to inhabit a world in which intellectual virtuosity can be purchased from a computer for pennies. Yet perhaps there is something healthy in the humiliation. We may discover that intelligence was a tool rather than the purpose of existence.

Mortality

There remains one characteristic of human beings that artificial intelligence may never share in quite the same fashion: we know that we are going to die. Whether machines can become conscious is unknown. Whether conscious machines could meaningfully experience mortality is still more obscure. Software can in principle be copied, restored and instantiated elsewhere. Human beings cannot.

Our mortality structures human meaning. We value childhood because it ends, relationships because people can be lost and opportunities because they disappear. Decisions matter partly because one cannot live every possible life. An immortal intelligence possessing unlimited time might understand this proposition perfectly. Whether it could experience its significance is another question.

The human condition is therefore not merely a defective approximation to machine intelligence. It is a particular form of existence, defined by embodiment, vulnerability, dependence upon others and finite time. Those limitations generate suffering. They also generate meaning.

What are human beings for?

We can now return to our deliberately improper question. If artificial intelligence eventually writes better legal opinions, discovers better medicines, designs better aircraft, predicts weather more accurately, composes beautiful music and explains philosophy more clearly than human beings can, then perhaps we should allow it to do these things. Human beings need not defeat machines in an intelligence competition any more than we need defeat excavators in a digging competition.

Our task would instead be to decide what to do with the extraordinary inheritance such machines might create. We could use artificial intelligence to reduce disease, eliminate tedious labour, expand knowledge and make material scarcity less oppressive. We could give people more time for families, friendship, creativity, travel, scholarship and public life. We could build societies in which employment ceased to be the principal measure of human worth.

Or we could construct a world of unprecedented inequality, algorithmic dependence and exquisitely personalised distraction. Artificial intelligence cannot make that choice for us without the choice itself becoming meaningless.

That may ultimately be the answer. Human beings are not here to be the universe’s fastest calculators, largest repositories of knowledge or most efficient producers of economic value. If another form of intelligence becomes better at those things, nothing essential about human dignity need disappear. The purpose of intelligence was always to help us live. We made the mistake of imagining that living was for the purpose of demonstrating our intelligence.

AGI, if and when that expression finally describes something we all agree has arrived, may strip away that confusion. The machines may become cleverer than us. They may eventually become cleverer than we can comprehend. They may solve problems that have defeated generations of scientists and perceive patterns in nature invisible to the unaided human mind. But no amount of machine intelligence can answer, on our behalf, the question of what sort of lives we wish to lead. That question was never computational. It was always human.