For many ages to come the old adam will be so strong in us that everybody will need to do some work if he is to be contented. We shall do more things for ourselves than is usual with the rich today, only too glad to have small duties and tasks and routines. But beyond this, we shall endeavor to spread the bread thin on the butter — to make what work there is still to be done to be as widely shared as possible. ... for three hours a day is quite enough to satisfy the old adam in most of us!

— John Maynard Keynes, “Economic Possibilities for Our Grandchildren” (1930)

A case in point is the latest generation of AI-based image-generation engines, which have given rise to conversations about the potential redundancy of artists and illustrators — a major category of creative work. Over the past few months, multiple tech companies have unveiled platforms like DALL-E 2, Midjourney and Stable Diffusion that all boast of easy-to-use, low-cost services that allow anyone to generate both photorealistic and artistic images based on any text they can dream up — and with striking results. Want to see “a robot hand that is sketching a picture of another robot in photorealistic style”? You got it! We used DALL-E 2 to create it for you on page 36. The system comes with lots of examples that are more original than our prompt, such as “a bowl of soup that is a portal to another dimension.”

What this will mean for the human workers who operate in this space remains to be seen. One firm pitted their flesh-and-blood graphic designer against DALL-E 2 in a series of John Henry-style challenges. In blind judging, the human won two of the three contests, but it took the human three days to render the three images while DALL-E completed the same tasks in a matter of minutes.

Some artists see these image-generation platforms as a way to free them for other aspects of creativity. But, however they choose to rationalize the arrival of AI, there is little debate that these new tools will reduce the amount of labor required to complete graphic illustrations — even if humans remain involved in the process for the next few years.

Of course, this application is only one of a number of technological advances in recent years that have reduced the need for labor in areas as varied as reading mammograms and ringing up grocery purchases. And the range of tasks that AI systems can accomplish continues to grow beyond traditional rote computation into areas that involve creativity and even emotional intelligence. This flood of advances prompts questions whose answers were left to science fiction writers until quite recently. Will human labor be replaced by AI — and, if so, how do we prepare society for a future without work?

The Three Stages Leading Up to the End of Human Labor

The progressive inroads of ever more sophisticated AI would seem to be a given, but that does not automatically imply that AI will replace human labor. Technological advances in virtually every human endeavor have repeatedly given rise to concerns that machines will end up doing all the work, but those concerns have been proven wrong time and again.

In the past, rather than eliminating the need for human labor, new technologies have only shifted the tasks for which humans are employed. For example, before the great waves of mechanization in the 19th century, the majority of Americans worked in agriculture; today only 1.5 percent of the population remains on the farm. This sort of profound shift can be observed even in the span of a single lifetime: some 64 percent of all job titles that are used today did not even exist in 1940. And while the churning hasn’t been in everyone’s interest, average pay has risen more or less in tandem with productivity. As a result, technological progress has broadly served to raise standards of living.

There is, however, no guarantee that history will repeat itself. Far from it: there is reason to believe that future technological change will decouple productivity growth from the fate of workers and may ultimately lead to the end of human labor. We believe this transition is likely to proceed in three stages.

First, technological progress will shift from increasing worker productivity as well as increasing the demand for human labor — as it has over the past 250 years — to reducing demand for labor. There are already harbingers: the wages of workers (especially males) with lower levels of education have been declining for the past four decades. More recently, even some categories of college graduates have failed to gain from technological advances.

In the second stage, machines will be able to substitute for any type of human labor. Technology already makes it possible to eliminate supermarket cashiers and to read mammograms without the help of a radiologist. That is still a long way from doing all of the work that humans do today. However, there is no law of nature implying that computers will not eventually be able to perform any cognitive task that the brain can accomplish, including creative tasks that require coming up with new ideas and interpersonal tasks that require navigating a complex social environment.

The most advanced supercomputers already incorporate greater computing power than our best estimates for the human brain. The software, which would allow that computing power to execute a seemingly endless variety of human tasks, still lags behind. However, professionals who work in and around AI see human brain superiority coming to an end. The optimists (pessimists?) think this could happen within this decade. And a majority of AI researchers expect that artificial intelligence will be able to do everything a human can sometime in the 2060s.

However, simply because AI will be able to do everything as well as humans does not necessarily mean that the switch will be costeffective. The third stage begins when machines can perform all economically valuable tasks more cheaply than humans earning barely enough to subsist. This is a crucial threshold. For once crossed, it would mean that individuals could no longer survive based on their labor. The cost of computing today makes it unlikely that this point will arrive soon. But there are no obvious conceptual barriers to getting there, since both processing capacity and software continue to improve at a rapid rate.

Today, for better or worse, markets determine the returns to labor and the technology embodied in capital. How would — and should — society distribute income when labor has little or no value?

Income, Wages and Work

Strip down the issue to the fundamentals, and consider a worker with a given earnings potential and a given income from investments and transfers like Social Security and gifts from family. There are three possible outcomes for them in a market-driven economy.

Perish. When the market value of their labor plus other income is below subsistence.

Work. When the wages and other income combined is high enough to survive, they may still choose to work — either out of necessity or, if non-labor income is sufficiently high, out of choice.

Opt out. If basic needs can be met from non-labor income and wages are not enough to make it worthwhile to give up leisure, they may drop out of the labor force.

While this seems a formal statement of the obvious, it becomes interesting when we move from an individual to a whole society and assume that technology is progressively reducing the market value of labor. In that world, who should work how much — and who should decide?

To leave everyone with an incentive to work would increase competition for the shrinking pool of jobs, driving down wages and making it ever more difficult to survive on income from labor. Thus, from a utilitarian perspective — the greatest good for the greatest number of people — it makes sense to sever the link between income and work and, in one way or another, distribute income via social transfers or through sufficiently broad ownership of capital.

During the transition to zero-value labor it would be optimal (again, from a purely utilitarian perspective) to phase out those workers who are less productive first. Eventually, when it becomes cheaper to make all goods and services without labor, humanity will have reached its well-deserved retirement from the workplace.

But What if We Like to Work?

Thus far, we’ve only considered how to allocate labor and distribute income based on productivity. But for many, the rewards of work are more than just wages. These benefits include (among others) social connections, a sense of challenge, the pleasures of contributing to society, a way to occupy the day. On the other side of the coin, work can be boring, physically tiring and in some cases dangerous. Indeed, the overwhelming majority of workers — 85 percent worldwide and 65 percent in the United States — are disengaged from their work. And in light of this, it makes sense to consider the non-monetary benefits (and costs) of work as well as labor productivity when deciding which workers should be phased out first.

Will Markets Sort It Out?

So far, the labor choices that we have broadly outlined are consistent with individual freedom of choice — if, indeed, individuals can earn wages that correspond to their productivity at the margin and if the nonlabor income offered reflects society’s values as to who deserves what, the described behavior will naturally fall into place. Or to paraphrase Adam Smith, individuals who obtain the right level of income support would make efficient choices about working as if guided by an invisible hand.

Yet many discussions of a transition to a workless future propose that governments intervene in labor markets — for example, that governments should create or guarantee jobs, or that individuals should be made to work as a condition for obtaining government support. The implicit premise of such proposals in a world in which work is economically redundant is that the invisible hand is broken, i.e., that markets’ failure to deliver efficient outcomes justifies intervention.

Indeed, there are two distinct arguments to rationalize government intervention. One rests on what economists call “externalities”: when others gain or lose as a consequence of my actions. Examples include network effects, where my behavior enables other workers to build more valued social connections, or where work fosters greater political stability. There is also a potential for negative externalities — think traffic congestion and pollution that comes from millions of people getting into their cars and commuting to work.

The other argument is grounded in the less familiar term “internalities,” which arise when individuals make decisions that are not in their own best interest. This can go in either direction. For example, people may not sufficiently appreciate how the structure that work provides to daily life contributes to their sense of well-being. Conversely, workers may not sufficiently appreciate the risk that workaholism may eventually undermine their happiness because they give short shrift to their personal lives.

While these examples of externalities and internalities may in theory justify public intervention, we are unsure whether intervention would ever be wise in practice. One could easily imagine (at least in the United States) that public policy would be biased toward demanding work as the price of enjoying the dividends of machine productivity, but this may end up undermining the welfare gains that could be obtained from ever-more productive machines. As society becomes more accustomed to life without work, we believe that people will be able to find other — and superior — ways of fostering social connections and political stability as well as filling their days pleasantly. It would be a pity if technological advances allow humanity to enjoy a life of fulfillment in some combination of leisure and knowledge-seeking, but outdated views about the moral place of workforce individuals to perform tasks better done by machines.

That said, we do believe that public policy could be very helpful in the transition. If there is a future without human labor, it will not happen all at once. Machines will substitute for human labor in stages, and it would be desirable to make the transition as smooth as possible by leveling the impact of lower labor demand across the workforce. For example, the government might use financial incentives to encourage shorter workweeks and longer vacations, or reduce the age at which individuals can claim government retirement benefits.

Allocating Work and Income

In the world as we know it today, there are two main institutions that allocate work and redistribute income: markets and social insurance. Perfect markets and perfect social insurance would both protect individuals from the risk of losing out to technology. But, in reality, neither mechanism works well because they distort incentives.

As a result, we live in a market economy without insurance mechanisms for covering long-term risks such as the redundance of labor. If, as we anticipate, wages decline sharply under pressure from productive machines, it will no longer be possible for individuals to meet their basic material needs through work. When this happens, only those with substantial savings or guaranteed outside incomes will survive. And in the United States, that’s just about everybody except retirees and one percenters.

Social insurance systems provide a mechanism to redistribute income where markets for long-term private insurance fail. Whether it is through cash transfers or subsidized provision of goods and services, they could fill in the gap, mimicking what an optimal risk-sharing arrangement in the market would look like and rescuing unproductive labor from the dismal consequences of redundancy. Moreover, if labor is well on the way to becoming redundant, we do not have to be concerned about distorting incentives to work anymore.

Universal Basic Income

One much-discussed form of social insurance is a universal basic income (UBI), which gives every individual a monthly check sufficient to cover subsistence. As long as they must work during the transition to an all-machine economy, the UBI could supplement declining wages. In addition, with their basic subsistence needs met, individuals may choose to scale back their employment, for example, by working fewer hours. Instituting a UBI would presumably have a negative effect on the labor supply — though, in experiments, it generally has not — but that would actually be one of its goals in this context.

Not all UBIs are created equal. First, a proper UBI must be universal and needs to be distributed without a work requirement. Many social programs, such as the earnedincome tax credit, require that individuals participate in the labor force to receive cash benefits. But in a future in which machines substitute for human labor, positive work incentives would be counterproductive.

Second — and this is really ambitious — if we truly care about all of humanity, some form of the UBI would need to cover the world, not just the citizens of rich countries. After all, when machines can replace all human labor, the effects will eventually be felt everywhere.

In fact, there are reasons to believe workers in poorer countries will be harder hit by automation. As such, policies to blunt the worst effects of automation should reach beyond the boundaries of individual nations.

Inconceivable, You Say? Maybe. Impractical? Maybe Not.

Most of the world’s population lives in countries that could eliminate extreme poverty without digging very deep into their own GDP. And in a future context in which economies could not even aspire to create enough jobs that are sufficiently productive to cover subsistence, the political pressure for some variation of a UBI would presumably increase.

Consider a closely related issue. Labor is not only the main source of income for individuals, but also for governments: most rely heavily on taxing labor to raise revenue. And in part due to a worldwide race to the bottom in the taxation of capital, the long-term trend has been to decrease the tax burden on capital income and rely ever more on labor income taxation. But if automation makes all human labor redundant, governments will no longer be able to raise funds from taxes on labor. They will need to tax other factors of production — ideally ones that are in inelastic supply, like land and mineral resources, so that the tax system does not greatly distort prices. Alternatively, one might use taxes to offset existing price distortion. A prime candidate: taxing carbon emissions.

In for a Penny…

We may be some time from the point at which labor truly becomes redundant. But we would argue that we’ve already entered the transition phase. Low-skilled work is losing market value, and while high-skilled work may seem invulnerable to these technological forces, there is every reason to believe it is not. For one thing, constraints on investments in skills in places like India and China are fading, almost guaranteeing rapid increases in the supply of highly trained workers. For another, as noted earlier, artificial intelligence is making inroads in many skilled services — notably medicine and industrial design.

To get ahead of the curve, we propose that we need to start now to develop the institutions to distribute the gains from future advances in AI — whether through a UBI or through broadly distributed stock ownership — and to educate the public on the logic behind it. At present, there is no good reason for a wholesale replacement of the targeted social insurance system that we have in the U.S. and other advanced countries — if we did, the neediest and most vulnerable individuals would lose a significant fraction of the social support they currently receive, increasing poverty. However, we propose implementing a small UBI or distributed ownership program that automatically scales up as labor’s share in the economy declines.

Moreover, we propose to slowly unlink our current social insurance system from labor markets since continuing to force low-wage workers into the labor market in order to receive benefits such as the earned-income tax credit, health insurance or retirement pensions will become less and less desirable for society. It would be wasteful to force humans to remain in low-productivity jobs as autonomous machines become ever more productive.

If we manage the transition well, establishing institutions to distribute the abundant output in a world in which machines can create everything on their own, we have a chance of creating the human flourishing that was first glimpsed at the dawn of the industrial revolution.