The economic settlement we mistake for capitalism may have depended on something much simpler: capital needed us.
For most people, the most valuable economic asset they have ever owned is not a house, a pension or a portfolio of shares.
It is their capacity to work.
They sell that capacity repeatedly over a lifetime. Its value pays the mortgage, raises the children, funds retirement and turns workers into consumers.
Capital owners possess a different kind of asset: ownership of the productive system itself.
For most of the industrial age, these two forms of ownership existed in an uneasy but productive relationship. Capital owned the factories, machines and increasingly the corporations. Workers owned their labour.
Neither side could entirely dispense with the other.
That dependence created conflict. Marx built much of his analysis around it. Workers wanted higher wages and better conditions. Capital wanted lower costs and higher returns. Each side tried to capture more of the value they produced together.
But beneath the conflict was something easily overlooked.
Capital needed labour.
A factory without workers produced nothing.
That gave workers something more important than income. It gave them leverage.
They could organize. Strike. Negotiate. Vote. Demand pensions, healthcare, shorter working weeks, safer workplaces and a larger share of the prosperity their labour helped create.
Capital rarely surrendered these things willingly. It conceded them because workers possessed something capital required.
Out of that struggle emerged what we might call the industrial settlement.
It was never particularly stable or fair. It varied enormously between countries. It excluded many people. It periodically collapsed into strikes, depressions, political violence and war.
But eventually a rough bargain emerged across much of the developed world.
Capital would own most productive assets. Labour would provide the work. Productivity would increase. Wages would allow workers to consume much of what the system produced. Governments would tax both sides and provide some combination of infrastructure, education, healthcare and social insurance.
Workers were therefore simultaneously inputs into production and consumers of its output.
That closed the circle.
And for several decades after the Second World War, particularly in North America and Western Europe, the arrangement worked remarkably well.
It worked so well that we began to mistake it for capitalism itself.
It wasn’t.
It was a settlement produced by a particular balance of power.
The Bargain Starts to Move
Technology has always altered that balance.
Agricultural machinery reduced the number of people required to produce food. Industrial machinery replaced enormous quantities of physical labour. Computers eliminated armies of clerks, typists, bookkeepers and telephone operators.
But new industries appeared. New occupations emerged. Productivity created additional wealth and demand.
The familiar reassurance therefore has considerable historical evidence behind it:
Technology destroys jobs, but technology also creates jobs.
Usually it has.
The question is whether artificial intelligence belongs comfortably inside that historical pattern.
AI differs from many earlier technologies because it does not merely automate a particular physical process or narrow clerical function. It increasingly substitutes for capabilities spread across an enormous range of occupations: writing, programming, analysis, research, design, translation, customer service, administration, diagnosis, planning and increasingly decision-making itself.
The important question isn’t whether AI will eliminate work.
There will always be things humans can do.
The important question is whether AI will eliminate enough of capital’s dependence on human labour to break the settlement built around that dependence.
That is a very different question.
Marx’s Missing Scenario
Marx imagined a world of increasingly concentrated capital and increasingly exploited labour.
What he did not seriously contemplate was capital becoming progressively less interested in exploiting labour because it no longer needed very much of it.
The distinction matters.
In the classical conflict, the capitalist wants the worker’s labour but wants to pay as little as possible for it.
The worker therefore retains bargaining power because withholding labour imposes a cost upon capital.
But what happens when the machine can increasingly perform the work instead?
The conflict changes.
It is no longer simply:
How should the proceeds of production be divided between capital and labour?
It becomes:
What counterbalances capital when capital no longer needs labour enough for labour to counterbalance it?
That may be one of the central political-economic questions of the AI age.
And we don’t have an answer.
Nobody Has to Decide This
It is tempting to turn this into a story about Silicon Valley.
A handful of extraordinarily wealthy technology executives are building systems that may eliminate millions of jobs. Their companies own the models, infrastructure and computing capacity. Their incentives are obvious.
It looks almost conspiratorial.
It isn’t necessary for it to be.
Imagine a company discovers that AI allows 100 people to perform work previously requiring 150.
The CEO does not need to hate workers.
The shareholders do not need to desire mass unemployment.
The board does not need to meet secretly and decide to weaken labour.
The company merely needs to behave rationally.
Its competitors face the same choice. If one lowers its costs and the others do not, the others become less competitive. Investors reward the businesses that improve productivity. Managers adopt the tools that allow them to produce more with fewer resources.
Millions of individually reasonable decisions begin pointing in the same direction.
The architecture produces the outcome.
This is not a conspiracy.
It is an optimization.
And that makes it considerably harder to stop.
The Asset Problem
This also changes how we should think about inequality.
For most people, their capacity to work has been their principal productive asset.
For wealthy households, ownership of capital is much more significant: shares, businesses, property, funds and other claims on productive assets.
Artificial intelligence potentially changes the relative value of those two things.
If AI makes capital dramatically more productive while simultaneously making some human capabilities less scarce, then the value of capital rises relative to the value of labour.
Nobody has to choose greater inequality.
Nobody has to design it.
It can emerge from the changing relative value of the assets people already own.
That is a harder problem than greed.
Greed can be taxed.
Monopolies can be broken up.
Campaign contributions can be regulated.
Corporations can be subjected to stronger rules.
But suppose all of that succeeds.
Break five enormous technology companies into fifty smaller ones.
Those fifty companies still have an incentive to automate.
Indeed, greater competition may make that incentive stronger.
Antitrust can reduce the concentration of capital.
It cannot restore capital’s dependence on labour.
The Old Answers
This is where much of contemporary political debate begins to feel strangely historical.
The traditional left proposes stronger unions, progressive taxation, universal services and greater redistribution.
The traditional right proposes lower taxes, deregulation, entrepreneurship and economic growth.
Both sets of arguments assume that the underlying machine remains broadly intact.
Workers work.
Companies employ them.
Consumers consume.
Governments tax the resulting activity.
Politics decides how the proceeds should be divided.
But what if the operating system has changed?
A union derives its bargaining power from the ability of workers collectively to withhold something an employer needs.
What happens when the employer needs progressively fewer workers?
A minimum wage determines what someone must be paid when employed.
It cannot require their employment.
Retraining works when displaced workers can acquire skills demanded elsewhere.
But if increasingly capable machines are simultaneously entering those new occupations, retraining risks becoming a race between human education and machine capability.
None of this makes unions, minimum wages or education pointless.
It means they were designed inside a particular settlement.
And 1954 is not an available destination.
Redistribution Is Not Power
There are more ambitious proposals.
Universal basic income.
Citizen wealth funds.
Universal services.
Robot taxes.
Higher taxation of capital.
Broader ownership of shares.
Some combination of these may eventually become necessary.
They address a genuine problem: if fewer people receive sufficient income through work, purchasing power must reach them through some other mechanism or the consumption side of the economy begins to break.
After all, capital faces an awkward problem of its own.
Workers were also consumers.
If companies progressively remove workers from production, who buys what the machines produce?
Redistribution can potentially close that loop.
Tax some of the returns to capital, transfer purchasing power to citizens, and consumption continues.
But something important has changed.
Income is not power.
A worker participating in production possesses a claim on the system because the system needs something from them.
A citizen receiving redistributed income possesses a claim because the political system has decided they should receive it.
Those are not equivalent relationships.
Redistribution can change who receives the returns from the system.
It does not necessarily change who owns the system that generates them.
A society in which productive assets are concentrated among a relatively small ownership class while everyone else receives sufficient income through redistribution might be prosperous.
It might even be humane.
But it would represent a profoundly different political economy.
The old settlement would be gone.
Perhaps Ownership Is the Question
That suggests that the eventual debate may not really be about jobs.
It may be about ownership.
If artificial intelligence becomes an increasingly important form of productive capital, and if productive capital captures an increasing share of economic output, then distributing income after production may be less consequential than distributing ownership before production.
Pension funds already provide a primitive version of this. Workers indirectly own portions of productive capital.
Sovereign wealth funds provide another.
Employee ownership, citizen funds, public investment vehicles and other structures could broaden participation in the returns generated by increasingly automated production.
None provides an obvious answer.
And ownership itself does not solve the separate human questions of purpose, status and meaning that arise when work becomes less central. Those deserve another essay.
But economically, broader ownership at least addresses the underlying change in the asset relationship rather than attempting indefinitely to compensate for it afterwards.
Creative Destruction Without a Destination
Capital’s enthusiasm for artificial intelligence is entirely understandable.
AI promises extraordinary productivity.
It may accelerate scientific discovery, improve medicine, reduce drudgery, expand human capability and create forms of abundance that previous generations could scarcely imagine.
Companies would be irrational not to explore it.
Countries may eventually conclude they cannot afford not to.
That is precisely the problem.
We’re watching capital enthusiastically dismantle a settlement without having any idea what comes after it.
No conspiracy is required.
No oligarch needs to wake up wanting to become an oligarch.
No committee needs to design the destination.
Each participant merely follows the incentives immediately in front of them.
And perhaps that is why our existing politics has so little to say about what is happening.
The right still talks about growth.
The left still talks about redistribution.
Populists talk about elites.
Technologists talk about abundance.
Unions talk about protecting jobs.
Almost everyone is trying to negotiate the terms of a settlement whose underlying source of bargaining power may itself be disappearing.
For two centuries, labour could demand a share of economic production because labour was necessary to economic production.
Artificial intelligence does not need to eliminate human work entirely to change that relationship.
It only needs to make human labour sufficiently less necessary.
And if that happens, the defining economic question of this century may not be the one Marx asked:
Who owns the means of production?
It may be the question he never had reason to ask:
What happens when the means of production no longer need you?