Who Can Afford the Future

The appointment exists. The person does not get there.

That gap can contain a bus journey, time away from work, a second booking, a language barrier, a fee or a task that is easy to postpone until postponement becomes neglect. A service can be available in a directory and absent from a person’s practical life.

One of the most useful things a technology can do is shorten that gap.

In the ACCESS randomized trial, young people with diabetes were offered either an autonomous AI eye-screening service at the point of care or the comparison pathway of referral and education. The trial took place at two sites of Johns Hopkins Pediatric Diabetes Center. Within six months, screening was completed by all 81 people in the analyzed AI group, compared with 18 of 82 in the analyzed control group, about 22 percent.¹

The intervention combined AI with a more convenient workflow. It measured screening completion, not prevention of blindness. Its lesson is narrower than a universal claim about healthcare, but it is substantial: the way help is delivered can change whether people receive it.

Eye-screening completion within six months in ACCESS. Point-of-care AI: 100%; referral and education: 22%. The comparison includes the delivery workflow, and concerns screening access rather than long-term health.

A model can be highly accurate and still leave a person at home with an unfinished referral. A service that brings the test into an encounter already taking place may remove a barrier. To understand the improvement, we have to look at the entire path through care.

That is the kind of progress I want accelerated. Find the points where people fall out of a service. Test ways of closing the gaps. Expand the approaches that work, with the staff, support and safeguards needed to keep them working.

A society that calls this merely convenient has lost sight of what inconvenience can cost people with little spare time or money.

The price beside the price

When a company announces that a tool is affordable, it usually means the amount charged for access. The user may face a longer bill.

A reliable device costs money. Connectivity costs money. Learning how to use the tool takes time. Checking its answers requires skill. If the service works poorly in a user’s language or situation, the user may have to do more work to get the same result.

These are possible costs to measure, not reasons to assume that every new service excludes people. Some tools reduce them dramatically. Voice interaction may make a task easier for someone who finds typing difficult. Translation may help a person ask a question they could not otherwise ask. A clear explanation may make an unfamiliar process less frightening.

But the benefit must survive contact with the whole task.

Imagine a small business owner using an assistant to prepare a tender. The writing becomes easier. Then the owner discovers that the application requires financial records in a format the business cannot produce, an insurance policy it cannot afford and a reference from a previous contract it was never given the chance to win.

The assistant has helped with one barrier. The institution has preserved the others.

This is why access policy should begin with an actual journey through the service. Watch where people stop. Ask what they need to continue. Do not assume that a free account has completed the work of inclusion.

The same question applies to scientists and public institutions. A research group may be able to obtain a model but lack the computing resources, data or expertise needed to use it well. A school may have software and no time for teachers to evaluate it. A public office may have a license and no one empowered to change the procedure that makes the software ineffective.

The remedy is investment in usable capacity. Shared research facilities, training, secure data access and competent public procurement can widen the set of people who can do valuable work. Each program should be tested against what it enables, not the number of accounts it distributes.

There is a strong pro-acceleration case for this public capacity. If only the wealthiest institutions can investigate new tools, the questions asked will reflect a narrow set of needs. More people able to test and adapt systems means more chances to discover neglected uses and neglected failures.

The public should be able to contribute to the direction of progress, not simply purchase what the leading providers happen to offer.

From a customer to an owner

Affordable access is important. It leaves a deeper question unresolved.

Who owns the source of the income?

A worker can become much more productive with a tool while receiving little of the additional return. A citizen can help finance education, research and infrastructure while remaining only a customer of the firms built upon those foundations. A community can bear the disruption of change while waiting for someone else to decide whether it deserves compensation.

In The AI Dividend, I proposed a different starting point: ordinary people should hold durable claims on part of the wealth created by a more automated economy. The aim is to broaden ownership while preserving the incentive and freedom to build useful things.²

A wage pays for work. A dividend pays a share of a return to an owner. People can receive both. The point of a broader ownership system is to make the second channel less dependent on inherited wealth or employment at one unusually successful company.

This is not a prediction that jobs will vanish. The argument does not require that prediction. It begins with the possibility that productive power can grow while the bargaining position of some workers weakens, or that large gains can flow through ownership rather than wages. A society can prepare for that possibility without claiming to know the final number of jobs.

One part of my proposal is a portable worker capital account. Contributions would follow the worker across employers rather than disappear when a job ends. Holdings would be diversified, so that a person’s savings are not tied entirely to the same company that provides their paycheck.

That last point is practical. If an employer fails, a worker already faces the loss of income. A badly designed ownership plan can place the worker’s accumulated capital in the same path of danger. The aim is to spread claims, not give a vulnerable person a second version of the same risk.

A second part is a citizen wealth fund. It would hold assets on behalf of the public, with rules for investment, reinvestment and distributions. A person would have a claim through membership, including when they are caring for someone, studying, unable to work or between jobs.

The exact legal design would depend on the country. It would need decisions about eligibility, administration, taxes and interaction with existing services. A proposal is not a check already in the mail. It becomes credible when its funding and obligations are clear enough to be examined.

The central change is still simple. Instead of asking people to accept disruption in exchange for a general promise of future prosperity, give them a defined stake in the result.

What should fund the claim

A policy that taxes every use of AI as if adoption itself were an offense would work against the purpose of this book. It could burden a small firm trying to become more capable and make useful assistance harder to afford.

The funding approach in The AI Dividend distinguishes ordinary adoption and normal entrepreneurial returns from extraordinary rents, publicly enabled windfalls and control of scarce bottlenecks.³

An economic rent, in this discussion, is a return connected to scarce control or advantage beyond what is needed to keep a resource in its productive use. Identifying it in practice is difficult. A large profit may reflect a valuable invention, a temporary lead, monopoly power or a mixture. A policy cannot solve that difficulty by declaring every successful firm guilty.

It needs a defensible base, thresholds people can understand and enforcement strong enough to resist relabeling. It should protect small firms and low-margin uses. It should avoid taking so much from a risky success that investment in the next useful idea becomes unattractive. It should also resist the claim that every public return will destroy innovation, regardless of its size or design.

Public support can carry terms agreed in advance. If the public takes part of the risk, it can negotiate a return when the project succeeds. That might take the form of an ownership interest, a royalty or another defined claim suited to the funding arrangement. The terms should be visible before the money is committed, so that the public and the recipient know the bargain.

Such a policy would not establish that taxpayers own every downstream invention simply because public institutions exist. It would create specific claims through specific arrangements. General gratitude is easy to evade. A carefully written right is more useful.

There are real tradeoffs. A levy can be shifted partly to customers or workers. A funding rule can encourage firms to reorganize around it. A poorly governed fund can reward political allies. Asset returns can fall. A promised dividend can become an excuse to neglect healthcare, housing or education.

These are design obligations. They are reasons to build the policy well and test its effects, rather than reasons to leave ownership entirely as it is.

The comparison should include the costs of concentrated gains. If people see technological change improving everyone else’s options while reducing their own, appeals to national progress will sound hollow. Broad ownership can give them a material reason to support development. That is a political argument for the proposal, not a measured guarantee of public acceptance.

The better bargain is to make the success of useful technology something more people can share in without having to hope that an employer or a politician later chooses generosity.

A rule that moves before a crisis

One danger of waiting for harm to become politically undeniable is that the response arrives after people have used up their reserves.

A household can spend its savings before a legislature agrees that a transition is serious. A town can lose employers before a new program becomes available. The eventual payment may help, but it cannot return all the choices that disappeared during the wait.

This is why my dividend proposal includes an automatic element. A published rule would increase public capital contributions when a sustained measure of labor’s share falls below an agreed benchmark. A smoothing period would reduce the chance of reacting to a noisy quarter. Caps, recession protections and scheduled democratic review would limit sharp or poorly timed adjustments.⁴

This is a proposed rule, not an existing entitlement or a forecast of what it would pay.

The important idea is that society can decide in advance how it intends to respond to certain changes. Automaticity can make help less dependent on a fresh political struggle each time a signal worsens. It can also create a new danger if the signal is badly chosen.

Labor’s share is not a complete measure of every worker’s position. A national average can hide differences across regions and groups. A decline can have several causes. The rule therefore needs public reasoning about what it measures, how it can be revised and which other evidence should inform the response.

A formula should make a decision more visible. It should not make the decision immune to criticism.

The same care applies to payouts. Spending all of a fund’s return may leave little protection against bad years or future needs. Saving everything can turn a public promise into something people never experience. The policy has to decide how much to distribute, how much to retain and how to explain that choice.

A statement people can read is part of the design. What assets are held? What fees were paid? What came in? What was distributed? What remains? Who checked the account? A fund with an impressive name and unreadable finances has not created meaningful public ownership.

Nor should people have to surrender their political judgment because a dividend arrives. Shared returns cannot purchase consent to surveillance, unsafe deployment or monopoly. A person can own a stake and still insist that the enterprise behave properly.

The aim is greater independence, including the independence to say no.

Access that can survive a disagreement

A tool can be affordable today and still place its users in a weak position tomorrow.

Suppose a school builds its lessons around one service. Its materials are stored there, its teachers have learned the interface and its pupils depend on the format. If the provider raises prices or changes terms, the school’s freedom to choose another system depends on how difficult it is to move.

This is a reason to care about portability and compatibility. People should be able to retrieve their own work in usable forms. Public buyers should ask what happens when a contract ends. Institutions should avoid making a necessary service depend on an arrangement from which exit is only theoretical.

Open models and open tools can help some users inspect, adapt and maintain systems. They can also require expertise and computing resources that many people lack. A downloadable model is not the same thing as an accessible service. At the same time, a polished closed service can be easy to use while leaving its users unable to inspect important decisions or continue independently.

The public interest cannot be read directly from either label. It depends on what people can actually do, what dangers a release creates and which forms of control they retain.

We should support a varied field: capable public research, responsible commercial providers, shared infrastructure and independent institutions able to test claims. This reduces the chance that one company becomes the only place where a useful future can be obtained. It also gives government alternatives to either dependence or a heavy-handed attempt to run everything itself.

Such variety is a form of practical freedom. A worker can try a different tool. A researcher can challenge a result. A public office can change suppliers. A community can compare terms. The existence of an alternative turns a request into a negotiation.

This is where the subtitle of this book becomes demanding. Progress that belongs to everyone cannot mean that everyone receives the same advertisement. It must mean that more people can use the help, shape the terms and share in the returns.

The ACCESS trial began with a concrete gap between a referral and a completed screening. Ownership policy begins with another gap, between producing greater wealth and having a secure claim on it. Both ask us to follow the promise until it reaches a person.

The first gap may be closed by a better service. The second requires rules about power and property. Neither should be postponed until after the future has been built around those who already possess the easiest access.

Build the tools. Build the route to them. Build the claims that let people stand inside the prosperity they help make possible.

The table

The table had been built from common wood,
But only seven names were carved in gold;
The rest stood back and watched the owners’ food.

The room was warm. The hands outside were cold.
They had supplied the timber and the ground,
And heard the feast was theirs when they grew old.

A child beneath the window heard the sound
Of plates set down, of laughter through the wall,
And asked which side of wealth her life had found.

We moved the table to the public hall.
We kept the names of those who built it there,
Then made a place where every child could call

A portion of its strength a rightful share.
The feast would need more planting in the rain;
The wood would need a carpenter’s repair.

We knew that giving once would not sustain
A people needing more than sums to spend.
We wrote the right to gather in the grain.

A chair was there for one they used to send
Back to the cold that waited at the door.
Her place became a right, not ours to lend.

The house was hers to help make something more.

Sources and notes