Imagine making a complete inventory of everything human beings do for one another.

Driving a lorry. Diagnosing an illness. Writing software. Designing a building. Teaching a class. Preparing a set of accounts. Managing a warehouse. Negotiating a contract. Composing music. Caring for somebody at home. Deciding which experiments a laboratory should run next.

Now take each activity in turn and ask whether sufficiently capable AI and robotics could eventually perform it.

For most of the list, the honest answer is probably yes.

Some activities will be difficult or uneconomic to automate for a long time. Law and custom will preserve human involvement in others. We may continue to value a performance, a competition or a handmade object precisely because a person produced it. But these are fragile foundations for an economic future. They ask humanity to crowd into a shrinking collection of protected spaces while machines expand into everything else.

The search for something that only a human can do may be mistaken for a more fundamental reason.

It assumes the list is finished.

There is no final list of useful work

An inventory of present work is a static picture. It holds our aims constant, introduces increasingly capable machines and calculates what remains. Under those conditions, every successful automation can only subtract from the human share.

But life does not develop under static conditions.

Every substantial increase in our capabilities changes the circumstances in which people live. It removes constraints, creates expectations and exposes problems that were previously hidden behind more immediate ones. People begin to want things that once seemed impractical, extravagant or impossible. They also discover consequences nobody had reason to consider before the change occurred.

Suppose AI makes it possible to give every child a patient and highly capable personal tutor. The first objective may be to teach the existing curriculum more effectively. If that succeeds, it does not complete education. It changes the questions.

What should children learn when individual explanation is abundant? Which parts of education are really about acquiring information, and which are about confidence, cooperation, judgment or discovering what is worth caring about? What becomes possible when a school is no longer organised around the limits of one teacher addressing thirty pupils at once?

Solving one constraint reveals a larger design space.

The same pattern could appear in medicine. Better diagnosis would allow more attention to move towards prevention, individual variation and quality of life. Successful treatments would change the population and create new medical and social questions. In business, removing routine administration might make levels of personalisation or service economical that previously required a much larger organisation.

Capability does not merely satisfy desire. It helps form desire.

This is why a photograph of today’s work is not a forecast of tomorrow’s purposes. If machines eventually become able to perform nearly everything on the inventory, humanity will not still be standing at the point where the inventory was written. We will be living with the consequences of everything that has already changed. We will know more, expect more and be able to contemplate ambitions that are not yet credible.

The world after a solution is not the world before it.

Not the last human advantage

This argument does not require us to nominate a mental ability that machines will never acquire.

AI may become better than people at proposing scientific questions, detecting unmet needs, designing institutions and anticipating the effects of a decision. It may generate ambitions that no person would have conceived. If the frontier of useful work expands, artificial intelligence can help expand it.

Nor should we quietly reserve the phrase working out what matters for humans while allowing machines to do everything else. Determining what matters involves observation, imagination and reasoning, all of which AI may perform extremely well.

But capability is not the same as standing.

People are not merely one source of proposals competing with a more intelligent source. We are participants in the world those proposals would change. We experience whether a new system is liberating or oppressive, whether an efficiency has removed drudgery or destroyed something important, whether a measured improvement corresponds to a life that actually feels better.

An AI may understand those responses, perhaps eventually better than any individual person. It may reveal preferences we struggle to articulate and consequences we have failed to anticipate. But understanding somebody’s interests does not by itself confer sole authority to decide between them.

There is a difference between working out how to achieve an objective and deciding which objectives should govern a shared world. The second question contains disagreements that greater intelligence cannot simply dissolve. A faster journey may require more surveillance. A more efficient care system may offer less continuity. A safer public space may be less private. The facts matter, but the choice also depends on whose interests count, which risks people may impose on one another and what sort of society they are entitled to help shape.

AI can reason about all of this. People can reason badly about it. The case for human involvement is not that our judgment is uniquely infallible. It is that the people who live with a decision have a legitimate place in making it.

Their experience also changes as capability changes. New systems create new habits, expectations, disappointments and possibilities. These become fresh evidence about what should happen next.

Knowledge from the world as it is lived

There is also a more immediate human contribution. Long before we confront the furthest possibilities of advanced AI, people possess an enormous stock of knowledge that could make present systems more useful.

Much of it is not written down. It has been acquired through years of contact with particular customers, patients, buildings, machines, organisations and communities. It appears as an exception noticed early, a question asked at the right moment or a suspicion that a sensible-looking result will fail in practice.

Consider an experienced coordinator arranging home-care visits. A scheduling system might initially optimise for punctuality, travel time and cost. The coordinator knows that, for some clients, repeatedly seeing the same small group of carers is not a sentimental extra. Familiarity reduces distress, makes subtle changes easier to notice and affects whether the care succeeds at all.

This is not a task that must remain beyond AI. Once the pattern is visible, a capable system may incorporate continuity, learn when it matters and optimise the rota more effectively than the coordinator could. It may later discover further patterns without being taught.

The important event is that somebody encountered reality and revealed that the original definition of a good schedule was incomplete.

At present, knowledge like this is usually sold together with a person’s time. An organisation pays for the coordinator to be present, and experience arrives as part of the labour. AI creates the possibility of separating the contribution from its repeated application.

The coordinator might demonstrate cases, compare good and bad results, correct proposed decisions and explain exceptions as they arise. They need not produce a perfect manual or decision tree. Conversation, observation and correction can draw out knowledge that would be difficult to formalise on demand.

Once expressed and tested, the insight could improve thousands of decisions the contributor never personally makes.

That separation raises an economic question. If a system repeatedly creates value using something a person taught it, should all of that value belong to the owner of the system?

Paying for contribution rather than repetition

For most of history, most people had little opportunity to turn practical insight into a durable economic asset. A small number wrote books, secured patents or built companies. Most knowledge remained local. It was passed to a few colleagues, applied manually throughout a career and often lost when the person left.

AI could change the economic shape of that knowledge. Someone develops a better way to recognise a failing project, prevent a manufacturing defect, assess a tender or organise a service. They teach the method to a system. The system applies it repeatedly. The contributor no longer has to perform every repetition for their original work to remain useful.

We already accept this shape in other areas. A songwriter need not perform a song personally every time it earns money. An inventor need not manufacture every product that uses an invention. These comparisons do not provide a ready-made legal model for practical knowledge, but they demonstrate the underlying possibility: an original contribution can remain connected to some of its later economic value.

That does not mean every observation should become private property or every preference should generate a royalty. Some knowledge should be a public good. Decisions about shared objectives often belong to democratic governance rather than a marketplace. Employees, collaborators and employers will have legitimate competing claims. Many useful insights will be independently discovered by several people.

But when a private system earns money by repeatedly applying a distinctive and identifiable contribution, paying the person who supplied it is not an absurd idea. It may be both fair and useful. Fair, because automation should not make the contribution disappear merely by making it repeatable. Useful, because people have more reason to share, test and maintain their knowledge when they participate in the value it creates.

This would require more than a promise. Systems would need provenance: a credible account of who contributed a method, how it changed and which later results still depended upon it. They would need ways to measure use and performance without rewarding confident nonsense. Terms would have to govern whether examples and corrections could be used to reproduce a contributor’s method and then remove them from the economics.

Payment could not necessarily continue forever. A contribution may become common knowledge, be independently discovered or matter less as subsequent improvements accumulate. Attribution should follow meaningful contribution, not create a permanent toll on every distant descendant of an idea.

These are difficult institutional questions. Leaving them unanswered, however, is also a choice. It allows whoever owns the system to absorb the knowledge and decide unilaterally what, if anything, its source receives.

Does the well run dry?

There is a serious objection to this picture.

Perhaps humanity has accumulated a valuable but finite stock of practical knowledge. People can teach it to machines and may be paid while that transfer occurs. But once AI has absorbed the lessons, improved upon them and learned to discover new methods for itself, why would it continue to need human contribution?

If the world and its objectives were fixed, this might be the end of the story. We would be extracting a reservoir of past experience. The better the process worked, the sooner the reservoir would be exhausted.

But practical knowledge is not only an inherited stock. It is also a flow produced by change.

Every successful application of intelligence alters the conditions in which the next decision will be made. A care system that solves continuity creates capacity to ask what a richer and more independent life at home might involve. An educational system that makes tutoring abundant reveals choices about the purposes and structure of education. A scientific advance makes new experiments possible and exposes new uncertainties. Greater prosperity changes what people regard as tolerable, necessary or worth attempting.

People experience these altered conditions. They form new preferences, notice new failures and imagine new aims. AI observes them too, finds patterns and proposes responses. Together—or sometimes separately—people and machines turn those responses into new methods. The methods are applied at scale, changing the conditions again.

The cycle is not sustained by a permanent defect in machine intelligence. It is sustained by the fact that improvement changes its own subject.

There may never be a final catalogue of worthwhile objectives because every consequential achievement changes what can be wanted and attempted next. Material resources remain finite, and not every dimension can improve without limit. But the possible arrangements of health, knowledge, relationships, institutions, environments and experience may be effectively inexhaustible.

For most of history, we could identify more worthwhile problems than we had the intelligence, labour or institutional capacity to address. AI relaxes those constraints. It may not empty the field of useful ambition. It may reveal how much larger the field is.

Possibility is not a distribution system

None of this guarantees that everyone will have well-paid work, or that selling insights will provide a secure income for most people.

An open frontier of possible improvement can coexist with concentrated wealth and power. The owners of capable systems may capture most of the gains. People may possess valuable knowledge but lack the access, bargaining power or legal protection needed to benefit from it. A platform for paid contributions could reproduce the familiar shape of the creator economy: a small number earning substantial sums, a long tail earning almost nothing and an intermediary capturing the dependable value.

Transitions may also be brutal. New opportunities will not necessarily appear in the same places, at the same time or for the same people whose livelihoods disappear. The possibility of contributing to humanity’s next ambitions is not an adequate response to somebody losing their income next year.

Payment for practical insight is therefore not a complete post-AI settlement. Broader ownership, public provision, social insurance and new distributions of technological wealth may all be required. Some of the most important human input—expressing needs, participating in collective decisions and reporting harm—should not depend on whether a market finds it profitable.

But contribution can still be one meaningful part of the economy. We should not preserve repetitive work merely to preserve employment. Nor should we accept an arrangement in which systems become more capable by absorbing people’s experience while treating the people themselves as disposable inputs.

A better principle is that when a person makes a demonstrable contribution to a useful system, they should have a credible route to recognition, influence and, where economic value is created, payment.

Beyond the shrinking residue

Predictions about AI often begin by surveying the present and subtracting everything machines may learn to do.

The result is understandably bleak. If civilisation contains a fixed quantity of useful thought and action, abundant machine intelligence must leave less room for human contribution. Hope survives only in tasks protected by technical difficulty, law or sentiment.

But civilisation is not a finished list of tasks. It is a continuing response to its own changing capabilities.

People have knowledge that can make AI more useful now. They can teach systems better methods and expose objectives that were incomplete. As those systems change the world, people will encounter new consequences and form new ambitions. AI will contribute its own discoveries, proposals and judgments. Each can make the other more capable, and every successful answer can change the question that follows.

The economic challenge is not to identify the final activity machines can never perform. It is to build institutions that recognise valuable contribution in a world where execution is increasingly automated, knowledge can be made durable and the frontier of worthwhile action continues to move.

The important questions are therefore not only What can machines do? or What is left for people?

They are: What has become possible? What would make the changed world better? Who helped us discover that—and how should they share in the value?

There is no final answer, because every good answer changes the world in which we ask.