Skip to content

Affect policy, ignore parties

Strategic Liberty Coalition
Log in

Think tank

Essay

Abundance Without Exit: Why Automation and AI Don't Make the Case for Socialism

Automation and AI won't deliver socialism's abundance. Without the freedom to exit, refuse the plan and compete, they build a smarter cage.

If machines can make almost everything, do we still need markets? A popular strand of the left says no: hand the robots to the state and let it provide. Follow that argument one step further and it leads somewhere its authors don't want to go.

As machines, software and now large-scale AI cut the human labor it takes to produce food, energy, goods and many services, a growing number of leftists argue that government should simply provide for the public's basic needs. The reasoning is tidy. If technology can meet material wants with less and less human work, then scarcity is becoming a political choice rather than a physical fact. The just response, they conclude, is public ownership of the productive machinery and a planned distribution of the surplus: universal basic income as a halfway house, universal basic services as the real thing, and "fully automated luxury communism" as the advertised destination.

The read on the technology isn't foolish. The political conclusion doesn't follow. In fact, if you take their own story about automation seriously and ask what kind of society can keep improving the machines rather than merely occupying them, the argument points the other way. A highly automated, AI-run economy that is also administered as a socialist state isn't a leisure commons. It's a system in which the last scarce goods (permission, coordination and the correction of mistakes) are monopolized by whoever controls the models and the plan. That isn't luxury. It's a more efficient cage.

We have made a related argument about the past. In "The Dictatorial Commune and the Chattel Plantation," we argued that systems which claim the producer's labor survive only by closing the exits. This essay applies the same test to the future. Abundance doesn't make freedom unnecessary. Without the freedom to leave, refuse and compete, abundance becomes one more thing the plan hands out.

What the Automation Argument Actually Claims

The lineage is real. In the Grundrisse, his 1857–58 notebooks, Marx sketched what is now called the "Fragment on Machines." Capital, he wrote, "presses to reduce labour time to a minimum." As industry develops, the creation of wealth comes to depend "less on labour time" and more "on the general state of science and on the progress of technology." The worker "steps to the side of the production process instead of being its chief actor." Machines are "the power of knowledge, objectified," and the extent of fixed capital shows how far social life has "come under the control of the general intellect" (Marx, Grundrisse, Notebook VII).

A modern literature has grown from that fragment. Paul Mason's PostCapitalism (2015) built on it directly: information goods cost almost nothing to copy, so "its price will fall towards zero," and Mason called for a "Project Zero" aimed at, among other things, cutting "necessary work time as close as possible to zero" (Mason, The Guardian, 2015). Aaron Bastani's Fully Automated Luxury Communism (Verso, 2019) is the popular statement. Bastani argues that renewables, automation, lab-grown food and even asteroid mining are producing "extreme supply," that capitalism still organizes itself around scarcity, and that the technology should be put to work for everyone rather than for private profit (Quillette review; Guardian review). His preferred policy isn't a cash basic income but "universal basic services": housing, transport, education, health care and information, free at the point of use (Socialism Today). The cash version comes from Nick Srnicek and Alex Williams, whose Inventing the Future (2015) makes four demands: full automation, a shorter working week, a universal basic income and the "diminishment of the work ethic" (Srnicek and Williams, interview).

More orthodox Marxists are suspicious of basic income. Writing in International Socialism, Ellen Clifford and Mark Dunk warn that a UBI would function "as a wage subsidy" and argue that "the fundamental issue with automation is not the need to replace income for workless humans but the question of the ownership of the technology itself" (International Socialism, 2020). Lenin's old phrase for the goal still fits: the state must hold the "commanding heights" of production (Lenin, 1922). Plan for need, not profit, then speed up automation under public control so that the remaining work can be shared and necessities provided free.

The factions differ, but they share a premise: technology is making the old link between work and income obsolete, and politics must catch up by socializing the machines.

Grant the technical half for the sake of argument. The labor needed to make many goods is falling. Software copies at almost no cost. Robots and AI models will keep taking over routine thinking and many kinds of physical work. None of that settles who should own the frontier, who is competent to improve it, or what happens when a political monopoly is bolted onto systems that already concentrate control.

The First Contradiction: You Can't Seize a Research Program

Nineteenth-century expropriation had a physical picture: the mill, the mine, the rail yard. Workers already ran the machines day to day. Capitalists held the title. Take the title and the mill keeps spinning, at least for a while.

Advanced automation isn't a mill. The "means of production" that matter now are model weights, data pipelines, software toolchains, chip fabrication measured in nanometers, energy systems, and the hard-won know-how of the people who can make the next version better than the last. These aren't objects a class can occupy. They're moving research programs. Most of the people who would be invited, in the old vocabulary, to "seize" them couldn't train the systems, debug the fabs or improve the algorithms. That isn't an insult to working people. It's a description of specialization. A nurse, a warehouse picker and a line cook aren't thereby qualified to set the training objective of an AI model or redesign a lithography line, any more than an AI researcher is qualified to run an ICU. Treating expropriation as if it transferred skill along with title is cargo-cult industrialism: capture the building and assume the capability comes with the keys.

The Marxist reply is that workers would hire, or become, the experts, under democratic control. That just moves the problem. Either the experts keep real discretion, in which case you have a new technical elite administering "the people's" capital (a nomenklatura with GPUs), or research and deployment answer to committees that don't understand the field. In the first case, "workers' control" is a slogan pasted over a hierarchy. In the second, the system degrades. There is no third option in which collective ownership magically transfers the unwritten knowledge that actually moves the frontier.

That's the first contradiction. The same writers who treat automation as powerful enough to abolish scarcity also treat it as simple enough that people who didn't build it can take it over without losing output. If the technology is powerful because it embodies rare, cumulative, specialized knowledge, then handing the title to people who don't have that knowledge doesn't liberate the productive forces. It taxes them.

The Second Contradiction: Wanting the Fruit, Cutting Down the Tree

Automation doesn't arrive as a finished gift. It's a sequence of bets: money sunk into research that usually fails, companies that die, scientists who are wrong for a decade, supply chains that exist only because someone expected a return. Socialist expropriation attacks that sequence at the root. When the rewards of improvement are socialized, the people who know how to improve the system get a worse deal. They can still draw salaries. They can't capture much of the upside of being right, and they can't easily leave to try a rival approach when the plan is wrong.

Planned economies didn't fail for lack of engineers. The Soviet Union trained plenty. They failed, as Ludwig von Mises argued in 1920, because without market prices for the means of production, planners have no way to calculate which uses of scarce resources are worth it (Mises, "Economic Calculation in the Socialist Commonwealth"). F. A. Hayek added that the knowledge a planner would need "never exists in concentrated or integrated form but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess" (Hayek, "The Use of Knowledge in Society," 1945). Markets run that discovery process in parallel, with millions of people testing guesses against each other. AI doesn't repeal the problem. It can make a planner more confident while leaving him just as ignorant of what should be produced, for whom, at what quality, and at what hidden cost. A model will optimize the official objective. It won't tell you the official objective is a political fiction.

There's a more immediate brake, and it doesn't even require full expropriation. A politics built around protecting jobs as such will delay the very automation it cites as the reason the state should provide for everyone. Marx himself noted that machinery devalues the worker's labor, "hence the workers' struggle against machinery" (Grundrisse). That struggle is alive today:

  • After a strike in October 2024, East and Gulf Coast dockworkers won a six-year contract providing that "there shall be no fully-automated terminals developed and no fully-automated container handling equipment used" through 2030, along with a ban on remotely operated ship-to-shore cranes (ILA–USMX Master Contract, Art. XI).
  • In 2023, California's legislature passed a Teamsters-backed bill requiring a human operator in every autonomous vehicle over 10,000 pounds, 69–4 in the Assembly and 36–2 in the Senate. Gov. Newsom vetoed it (bill history; Reuters).

Workers protecting their livelihoods are acting rationally, and plenty of those votes came from both parties. Our point is about where the incentives lead once the state owns the machines. Under an ideology that ties dignity to employment, the pressure to freeze deployment gets stronger, not weaker. The Soviet constitution of 1977 guaranteed "the right to work" and in the same document declared that "evasion of socially useful work is incompatible with the principles of socialist society" (USSR Constitution, Arts. 40 and 60). In 1964 the poet Joseph Brodsky was sentenced to five years of internal exile at hard labor for "social parasitism," essentially for working as a poet instead of holding an approved job (Nobel Prize biography). You can't argue at once that machines should liberate humanity from work and that the state must preserve work so people remain politically recognizable as workers. One commitment has to win, and in our judgment, wherever the state is the employer of last resort, the job usually wins. The robot is delayed. The abundance that was supposed to fund the commons arrives later and thinner.

So the position is unstable on its own terms. It needs accelerating automation to make planning for need look cheap. Its theory of property and its theory of employment both slow that acceleration down. The predicted surplus then becomes the moral argument for a system that would have produced less of it.

Utopia on Paper, Administration in Fact

The utopian picture is familiar: necessary labor collapses, the remaining tasks are shared, goods are abundant, work and leisure blur into play, and the state, having delivered what Marx called "the true realm of freedom," withers away (Capital, Vol. III, ch. 48). Marx famously refused to write "recipes … for the cook-shops of the future" (Capital, Vol. I, 1873 Afterword). His modern heirs are less shy about the kitchen. Bastani describes 3D-printed steaks and asteroid mining in loving detail, but even a broadly sympathetic reviewer in The Guardian found that "much of the book is light on politics" (Guardian). The menu is specified. The offices that would run the restaurant aren't.

Now add the thing these visions treat as a neutral tool: an AI-run society. Allocation, credit, medical triage, energy dispatch, content ranking, policing of "misinformation," eligibility for housing and stipends: all of it becomes software. In a liberal order with exit, competing platforms and private property, a bad model is a company you can leave, a product you can refuse, an agency you can sue or vote out. In a socialist state that has claimed the means of production as public infrastructure, the model is the plan. Disagreeing with its outputs isn't consumer choice. It's noncompliance.

That's the dystopia the automation argument won't think through. A government that guarantees basic needs must define those needs. In a low-tech welfare state the definitions are crude and enforcement is leaky. In a high-tech socialist state the definitions can be granular and enforcement continuous. The system that "gives" you housing, food, bandwidth and mobility is the same system that can make them conditional on behavior, speech, carbon scores or political reliability. You don't need a cartoon villain. You need only the ordinary incentives of a monopoly administrator: cut measured waste, punish fraud, keep the plan stable, treat dissent as noise to be filtered out.

We don't have to imagine what automated administration of benefits looks like when it goes wrong. Liberal democracies have already run the experiment on a small scale:

  • Australia's "Robodebt." From 2016 the government used automated income averaging to issue welfare debt notices. The courts found the method unlawful, and a royal commission called it "a crude and cruel mechanism, neither fair nor legal" that "made many people feel like criminals" (The Guardian, 2023; Prime Minister of Australia).
  • The Dutch childcare-benefits scandal. The Dutch tax authority used a risk-scoring algorithm, with nationality among its risk factors, to flag benefit fraud (Amnesty International, "Xenophobic Machines"). As many as 26,000 parents were wrongly accused and told to repay money, and in January 2021 the government resigned (BBC).

Notice what fixed those failures: courts, a free press, opposition parties, elections, and citizens with lives and incomes outside the agency that wronged them. Those are exactly the checks that weaken when the state becomes the sole provider of everything. In our view, a system that runs every household's food, housing and work through one administrator, with no outside supplier and no exit, would make errors like these harder to notice and much harder to escape.

Central planning plus total information isn't the withering of the state. It's the state finally getting the nervous system it always wanted. Engels promised that under socialism "the government of persons is replaced by the administration of things," but the same passage begins with the step that makes it all possible: "The proletariat seizes political power and turns the means of production into State property" (Engels, Socialism: Utopian and Scientific). The state comes first. The withering is a promise. And an AI that assigns rations, shifts and permissions is still governing persons. It's just doing it at a resolution no commissariat ever managed.

Markets Correct Errors. Monopoly Plans Repeat Them.

The deeper point is about mistakes. Markets can be brutal and are often unfair, but they're a distributed error-correcting machine. Prices carry information about scarcity that no committee can see. Failure bankrupts a firm rather than a population. Socialist automation turns that upside down. When the plan is wrong, the model is retrained on the plan's own data, and the feedback becomes circular. The people who notice the mistake have no stake in fixing it and no legal alternative supplier. In a world of factories, the result is shortages and black markets. In a world of AI infrastructure, the result can be a smoothly administered wrong answer.

The Conclusion They Won't Draw

Follow the original premise without the romance. If technology really is driving the labor cost of subsistence toward zero, two things become true at once.

First, the question of a floor gets more pressing. A society that can feed and house people with less and less labor will face hard questions about the people who still fall through the cracks. Libertarians answer those questions in different ways. Milton Friedman proposed a negative income tax, and Hayek accepted a guaranteed minimum income outside the market (Quillette). Many others, especially anarchists and voluntaryists, would rather see abundance flow through charity, mutual aid, private dividends and cheaper goods than through any state check. Where all of them agree is that a floor, if there is one, shouldn't require the recipient to win a political game in order to eat.

Second, and more important, the case for limiting the institution that writes the checks gets stronger too. The more the material problem is solved by machines, the more the political problem becomes control of the machines. Abundance doesn't abolish power. It changes the currency of power from ownership of labor time to ownership of protocols. A society that responds by fusing ownership of those protocols with the state's monopoly on force has built the most complete hierarchy available to it.

That's why the libertarian conclusion isn't a slogan bolted on from outside. It's the same logic, carried one step further. If production no longer needs masses of commanded bodies, the excuse for commanding bodies shrinks. What stays valuable is the right to refuse a plan, to start a rival system, to keep the reward that pays for the next experiment, and to treat other adults as ends rather than as rows in an allocation table. Those are the institutions of a free society: property, contract, and exit, with as little coercive power as possible concentrated in any one stack. Markets are how people vote with their dollars every day, and the ability to take your business elsewhere is the vote that no plan can rig.

This Is Not a Defense of Big Tech

None of this requires pretending markets are kind, that today's tech companies are public-spirited, or that concentrated private AI is harmless. Private monopoly over foundational models is a real danger, and so is the cozy alliance between big firms and the governments that regulate them. The socialist answer, to make the monopoly official and call it the people, doesn't reduce that danger. It removes the last unregulated competitor and the last legal space to refuse.

This isn't only a debate on the left, either. Washington has been taking ownership stakes in strategic tech industries, including a 9.9 percent government stake in Intel in 2025 (Intel Form 8-K). We've argued that this kind of state ownership is a serious liberty cost that a single popular policy win doesn't cancel out (A Single Pro-2A Win Does Not Outweigh War, Tariffs, and State Ownership of Industry). The direction of travel matters more than the party label on it.

What This Means on Your Ballot

The future of automation won't be decided in a book review. It will be decided by votes on bills like the ones above: bans or mandates on autonomous vehicles, licensing rules for AI models, public equity stakes in chipmakers, and whether new benefits come with new surveillance. Those are policy questions, and the Strategic Liberty Coalition's Methodology is built to weigh policy, not party. Score how much harm a rule does to peaceful people, how many people it reaches, how much the office can actually change it, and whether a candidate who would move it toward liberty can win. Then use the free worksheet to run the races on your own ballot. Sometimes the candidate who moves the policy furthest toward liberty is a Libertarian, and sometimes it's someone else. Vote for the person who moves the policy.

The Bottom Line

The story of a working-class seizure survives as rhetoric because it flatters a nineteenth-century heroism the technology has made obsolete. You can occupy a plant. You can't occupy a capability you don't understand without making it worse. You can vote to share the output of a system that already exists. You can't vote the next system into existence if the vote destroys the process that discovers it. And you can imagine a post-work commons only if you refuse to notice that the administrator of an automated commons, armed with models that see everything the plan needs to see, has no structural reason to wither away.

The left that leans on automation is right that the old necessity of mass toil is weakening. It's wrong to think that weakening is a brief overture before a party that never leaves the stage. A society that becomes more technocratic without becoming more free doesn't graduate from scarcity into luxury. It graduates from scarcity into a more refined form of rule.

Sources

Marx, Engels and Lenin

The automation left and its critics

Economic calculation

Resisting automation; work as a duty

Automated administration gone wrong

State ownership in tech