The Marionette Fallacy
Why AI Will Not Replace Humanity—and Why the Proposal That It Will Is Neither New, Nor Honest, Nor Free
Keywords: artificial intelligence, automation, robotics, universal basic income, labour economics, creative destruction, technological unemployment, central planning, Austrian economics, comparative advantage
Thesis: The claim that artificial intelligence and robotics will render human labour obsolete—and that government-administered income transfers are therefore necessary—rests on a premise that is historically falsified, economically illiterate, and technologically uninformed. Every major wave of automation in the past three centuries has shifted labour rather than eliminated it, creating more occupations than it destroyed. Present-day robotics remains a controlled tool—a marionette—incapable of autonomous economic agency, and the proposal to pre-emptively replace market adaptation with state dependency is not a solution to a technological problem but the reassertion of an ideological preference that has failed wherever it has been tried.
I. The Keynesian Attic
Elon Musk recently offered the following proposition: “Universal HIGH INCOME via checks issued by the Federal government is the best way to deal with unemployment caused by AI. AI/robotics will produce goods & services far in excess of the increase in the money supply, so there will not be inflation.”
Strip the Silicon Valley packaging and what remains is a doctrine older than computing itself. In the 1930s and 1940s, the prophets of mechanisation made identical claims with identical confidence. Machines would produce abundance beyond measure. Labour would fade into irrelevance. Leisure would become the human condition. The only question was how to distribute the surplus—and the answer, then as now, was a central authority writing cheques.
Keynes himself, in his 1930 essay “Economic Possibilities for our Grandchildren,” predicted that by 2030 the economic problem would be solved and mankind’s challenge would be how to fill its idle hours. He coined the phrase “technological unemployment” not as a warning but as an inevitability to be managed. Nearly a century later, the economic problem persists, the working week has not vanished, and the only thing that has changed is the branding on the prediction.
The script is recycled. Replace “industrial machinery” with “AI,” swap “the dole” for “Universal High Income,” and the chorus resumes. Limitless goods. Effortless living. A benevolent state distributing stipends like a landlord permitting tenants to exist on his property.
The premise was wrong then. It is wrong now. And the prescription—dependency dressed as liberation—is not merely wrong but dangerous.
II. Three Centuries of the Same Prediction
The fear that machines will destroy work is not a modern anxiety. It is a recurring panic, and its track record is unblemished: it has been wrong every single time.
In 1589, William Lee invented the stocking frame knitting machine. Queen Elizabeth I refused him a patent, reportedly fearing it would deprive her subjects of employment. The textile industry proceeded to grow for four centuries, employing more people at higher wages with each successive wave of mechanisation.
The Luddites of 1811–1816 smashed looms in the English Midlands, convinced that power weaving would annihilate their livelihoods. By the middle of the nineteenth century, the British textile industry employed vastly more people than it had before the power loom existed. The machines had not destroyed work; they had created an entirely new industrial economy with occupations that could not have been conceived a generation earlier.
In 1964, a group of distinguished scientists, economists, and technologists sent a memorandum to President Lyndon Johnson—the “Ad Hoc Committee on the Triple Revolution”—warning that automation was creating a permanent class of unemployable workers and that the government must provide a guaranteed income. Within five years, unemployment had fallen to 3.5 per cent. The permanent class of the unemployable never materialised. What materialised instead was the service economy, the information economy, and eventually the digital economy—none of which had been predicted by the memorandum’s signatories.
In the 1990s, Jeremy Rifkin’s The End of Work declared that information technology would eliminate hundreds of millions of jobs worldwide. What followed was the longest peacetime economic expansion in American history, the creation of entirely new industries—web development, digital marketing, cybersecurity, mobile application design—and the lowest unemployment rates in decades.
The pattern is not subtle. It is not ambiguous. It is the most robust empirical regularity in economic history: technological displacement of labour in specific tasks is invariably accompanied by the creation of new tasks, new industries, and new forms of employment that were not—and could not have been—foreseen before the technology existed.
The reason for this is not mysterious. It is comparative advantage, operating at scale and over time.
III. The Economics That Musk Ignores
The argument that AI will make human labour obsolete rests on a confusion between absolute and comparative advantage—a confusion that David Ricardo resolved in 1817, and which apparently requires resolving again every generation.
Even if artificial intelligence could perform every task more efficiently than a human being—a claim that is itself absurd, as I will address—it would still not follow that humans have nothing to contribute. Ricardo demonstrated that trade is mutually beneficial whenever opportunity costs differ, regardless of absolute productivity. A barrister who types faster than his secretary still benefits from employing the secretary, because every hour spent typing is an hour not spent practising law. The relevant question is never “can the machine do this?” but “what is the opportunity cost of the machine doing this instead of that?”
When a technology automates one set of tasks, it does not eliminate the economic value of humans. It changes the relative value of different human activities. Tasks that are complementary to the technology—tasks that require judgment, creativity, social intelligence, contextual understanding, ethical reasoning, physical dexterity in unstructured environments—become more valuable, not less. The history of automation is the history of this revaluation.
The introduction of ATMs did not destroy bank teller employment. It reduced the cost of operating a branch, which led to more branches, which led to tellers being redeployed into relationship management and financial advisory roles—roles that required precisely the human qualities that ATMs could not replicate. The introduction of spreadsheet software did not destroy accounting. It destroyed the specific task of manual tabulation and created an explosion in financial analysis, modelling, and consulting—fields that could not have existed at their present scale without the automation of arithmetic.
This is not a pattern that AI disrupts. It is a pattern that AI continues.
Large language models are astonishingly good at pattern completion, text generation, and information synthesis within trained distributions. They are not good at navigating novel physical environments, making ethical judgments under genuine uncertainty, managing interpersonal conflict, repairing plumbing in a house built in 1947, or any of the thousand tasks that require embodied experience, contextual adaptation, and what Hayek called “knowledge of the particular circumstances of time and place.” The jobs that AI will create do not yet have names—just as “social media manager” and “cloud architect” and “UX researcher” did not have names in 1995. The inability to name them in advance is not evidence that they will not exist. It is evidence that the future is not centrally plannable.
IV. The Marionette: What Robotics Actually Is
The conversation about AI and labour is systematically distorted by a conflation of two very different things: software that processes language and data, and physical machines that act in the world. The word “robotics” is deployed as though we are on the threshold of autonomous mechanical agents replacing human workers across the economy. We are not. We are not close. And the people making these claims either do not understand the technology or do not care to represent it honestly.
A modern industrial robot is a marionette. It is a machine that operates under direct human control, executing pre-programmed sequences in highly structured environments. A robotic arm on an automobile assembly line performs the same welding operation, on the same joint, in the same position, thousands of times per day. It does this brilliantly. It does this because the environment has been engineered to accommodate the robot—not because the robot has adapted to the environment.
Move that robot three feet to the left, change the angle of the joint by fifteen degrees, introduce a part that is slightly out of specification, and the robot fails. It does not improvise. It does not adapt. It does not “figure it out.” It stops, or it produces defective work, and a human being intervenes to diagnose and correct the problem.
The Boston Dynamics videos that circulate on social media—robots doing backflips, navigating obstacle courses, dancing—are impressive engineering demonstrations. They are also carefully choreographed performances in controlled environments, executed after thousands of hours of programming, testing, and iteration by large teams of highly skilled engineers. The robot is not deciding to dance. It is executing a motion sequence that was designed, tested, debugged, and refined by humans. It is, in the most literal sense, a puppet—a sophisticated puppet, but a puppet nonetheless.
The gap between a robot performing a choreographed routine in a laboratory and a robot autonomously navigating an unfamiliar kitchen, identifying that the milk has gone off, deciding to make tea instead of coffee, and carrying a cup to a specific person while avoiding a child’s toys on the floor is not an engineering gap that will be closed by faster processors or more training data. It is a conceptual chasm that reflects fundamental limitations in how current AI systems understand—or rather, fail to understand—the physical world.
Current AI systems do not have world models in any meaningful sense. They have statistical correlations derived from training data. A large language model does not “understand” that water flows downhill; it has encountered the sentence “water flows downhill” enough times to predict it as a likely completion. The difference between statistical correlation and causal understanding is the difference between a parrot that says “fire” when it sees flames and a person who understands combustion. The parrot’s utterance is not knowledge. Neither is the model’s.
This matters because physical interaction with the real world requires causal understanding, not statistical prediction. A drone is not an autonomous agent; it is a remote-controlled device with stabilisation algorithms. A self-driving car—after decades of development and billions of dollars of investment—still cannot reliably navigate a construction zone or a parking lot with unusual markings. A warehouse robot operates in an environment that has been redesigned from the ground up to accommodate its limitations—standardised shelving, consistent lighting, uniform packaging, mapped floors.
None of this is autonomy. All of it is control. The robot is the puppet. The human is the puppeteer. And the string—the control, the programming, the structured environment, the constant human oversight—is what makes the performance possible.
When someone tells you that robots will replace all workers, ask them a simple question: can your robot change a nappy? Can it negotiate a fee dispute? Can it comfort a grieving widow? Can it tile a bathroom that is not quite square? Can it teach a six-year-old to read when the child is distracted, tired, and does not want to? The answer to every one of these questions is no—and these are not edge cases. They are the economy.
V. The Inflation Fallacy
Musk’s second claim deserves separate attention: “AI/robotics will produce goods & services far in excess of the increase in the money supply, so there will not be inflation.”
This is not economics. It is accounting dressed as economics, and bad accounting at that.
Inflation is not merely the ratio of money to goods. It is the consequence of monetary expansion interacting with the structure of production, the velocity of money, expectations, relative price changes, and the specific channels through which new money enters the economy. When a government issues cheques to every citizen, the new money does not arrive proportionally across all sectors simultaneously. It arrives where recipients spend it—on housing, food, energy, and services—sectors where AI-driven productivity gains are weakest and where supply constraints are most binding.
The result is not a general equilibrium in which rising output absorbs rising money supply. The result is Cantillon effects: those who receive the new money first—or who own assets that appreciate with monetary expansion—benefit at the expense of those further down the transmission chain. This is not theoretical speculation. It is the observed consequence of every major monetary expansion in recorded history, from John Law’s Mississippi Scheme to the post-2008 quantitative easing regime that inflated asset prices while real wages stagnated.
Moreover, the claim that AI will produce goods “far in excess” of the money supply increase assumes that AI productivity gains are both immediate and uniform across the economy. They are neither. Productivity gains from AI are concentrated in information-processing tasks—data analysis, content generation, code production, document processing. They are negligible in construction, plumbing, electrical work, elderly care, childcare, food preparation, agriculture in unstructured environments, and the vast physical economy that constitutes the majority of what people actually spend money on.
Printing money to distribute to a population while claiming that AI-generated abundance will absorb the inflationary pressure is not a policy. It is a fantasy. And it is a fantasy with a track record: it has been tried, under various names and justifications, and it has produced inflation, dependency, and the erosion of productive capacity every time.
VI. The Dependency Trap
Set aside the economics for a moment and consider what is actually being proposed. A central government issues income to every citizen. The income is not earned. It is not exchanged for productive contribution. It is distributed—by political decision, through political channels, subject to political conditions that may not be explicit today but will inevitably emerge tomorrow.
This is not liberation. It is the most complete form of dependency ever devised.
When your income derives from your own productive effort—whether as an employee, an entrepreneur, a tradesperson, or a freelancer—your economic existence is independent of political favour. You may be taxed, regulated, and constrained, but the fundamental source of your sustenance is your own capacity to create value that others voluntarily purchase. You have agency. You have bargaining power. You have the ability to walk away.
When your income derives from a government cheque, you have none of these things. Your economic existence is contingent on the continued willingness of the political apparatus to issue the cheque—and on its continued willingness to issue it to you. Today the cheque comes without conditions. Tomorrow it comes with conditions. You must live in this area. You must consume these products. You must hold these opinions—or at least not express those opinions. You must submit to this monitoring, this assessment, this review. The conditions will be introduced gradually, framed as accountability, and enforced as compliance.
This is not speculation. It is the observed trajectory of every welfare system in the history of the modern state. Programmes that begin as unconditional safety nets acquire conditions, requirements, surveillance mechanisms, and compliance regimes. They do so because the political logic is irresistible: if the state is giving you money, the state will eventually want to control what you do with it, how you live, and whether you deserve it. The notion that “Universal High Income” would be exempt from this dynamic is naïve to the point of wilful blindness.
Ludwig von Mises observed that interventionism is not a stable system—each intervention creates distortions that demand further intervention, in an expanding spiral that tends toward comprehensive control. A universal income programme administered by the federal government is not the end of interventionism. It is the beginning of a dependency architecture from which exit becomes progressively more difficult. Once a population derives its income from the state, the state’s leverage over that population is total. Every political priority, every ideological fashion, every electoral calculation can be embedded in the conditions attached to the income—and the recipients, having no alternative source of sustenance, will comply.
This is not freedom. It is serfdom with better branding.
VII. What Actually Happens
What actually happens when a transformative technology enters the economy is not what the prophets of obsolescence predict. It is messier, more creative, more human, and vastly more productive than any central planner could design.
The automobile did not merely replace the horse. It created suburbs, shopping centres, motels, drive-through restaurants, interstate highways, petroleum refining at industrial scale, traffic engineering, auto insurance, car dealerships, and an entire culture organised around personal mobility. None of this was planned. None of it was predicted. None of it could have been predicted. It emerged from millions of individuals making decisions, adapting to new possibilities, and creating value in ways that no committee could have anticipated.
The personal computer did not merely replace the typewriter. It created an entire digital economy—software development, IT support, web design, digital marketing, e-commerce, cybersecurity, data science, social media, streaming entertainment, the gig economy, remote work, and hundreds of other industries and occupations that did not exist and could not have been imagined in 1975.
The internet did not merely replace the post office. It created a global information economy of such scale and complexity that it defies comprehensive description. There are people today whose entire livelihood consists of activities that would have been unintelligible thirty years ago—managing online communities, optimising search engine rankings, designing user experiences for mobile applications, creating content for platforms that did not exist five years ago.
AI will follow this pattern. It will not eliminate human economic activity. It will transform it, in ways that are unpredictable, disruptive, and ultimately expansive. There will be displacement—there always is. Some jobs will cease to exist. Some industries will contract. Some skills will lose their market value. This is not a crisis. It is the ordinary operation of a dynamic economy, and it has been occurring continuously since the first human picked up a tool.
The appropriate response to displacement is not a government cheque. It is adaptation—retraining, entrepreneurship, migration to growing sectors, and the thousand forms of creative response that individuals undertake when they are free to do so and incentivised by the necessity of earning their own way. Markets do this. They do it imperfectly, unevenly, and sometimes painfully. But they do it—which is more than can be said for any centrally administered alternative.
The jobs that AI will create do not yet exist. They will emerge from the interaction of new capabilities with human creativity, in domains we cannot presently name. This is not a weakness of the argument. It is the argument. The inability to predict what a free economy will produce is not a justification for replacing it with a planned one—it is the reason planning fails. Hayek’s knowledge problem is not a rhetorical device. It is the fundamental constraint on all attempts to centrally direct economic activity, and it applies with full force to the project of deciding, from a government office, what 330 million people should do with their days and how much they should be paid for doing nothing.
VIII. The Real Danger
The real danger of the AI-will-replace-everyone narrative is not that it is true. It is that it is useful—useful to those who benefit from dependency, useful to those who prefer managed populations to free ones, and useful to those who have always believed, despite all evidence, that the right committee with the right algorithm could run an economy better than the distributed intelligence of a free market.
The narrative serves a political function. If people believe they are about to become economically superfluous, they will accept—even demand—the very dependency that would otherwise be intolerable to them. They will trade autonomy for security, agency for stability, and the productive dignity of earned income for the patronising comfort of a state allowance. And once they have made this trade, reversing it becomes nearly impossible—because dependency, once established, creates its own constituency, its own bureaucracy, and its own political inertia.
This is not compassion. It is control. And the fact that it is proposed by a billionaire technologist rather than a politician does not make it less so. The history of industrial paternalism—the company town, the factory village, the benevolent employer who provides housing, food, and recreation in exchange for docility—is not a history of liberation. It is a history of control dressed in the language of generosity.
Musk’s proposal is the company town scaled to a continent. The company is the federal government. The town is the nation. The currency is compliance. And the product being manufactured is not goods or services but a population that has been convinced it cannot produce anything of value—and has therefore surrendered the right to try.
IX. Discard the Premise
No. Discard the premise entirely.
AI will not replace humanity. Robotics is not autonomous—it is a marionette, a tool, a drone operated by human hands and human minds. The economy is not a fixed quantity of work to be divided among a fixed population. It is a living, evolving, expanding system of human cooperation, driven by creativity, adaptability, and the irreducible fact that human beings are not interchangeable units of production but unique agents with knowledge, judgment, and purpose that no algorithm replicates.
The proposal that the government should issue cheques because machines are about to make people obsolete is not a forward-looking policy. It is the same backward-looking fantasy that has been offered, and falsified, for three hundred years. It is the Luddites with a press release. It is Keynes without the elegance. It is central planning without the honesty to call itself by its proper name.
Every technological advance in human history has done the same thing: created disruption, created opportunity, and created the necessity for adaptation. AI will do the same. The appropriate response is not to pre-emptively surrender human agency to a government programme. It is to allow markets to function, individuals to adapt, and the unpredictable, unplannable, irreducibly human process of creative destruction to do what it has always done—produce a future that is richer, more complex, and more free than anything a committee could have designed.
The rest is fantasy—expensive, coercive, and all too familiar.