Trump, in a gathering of AI overlords welcomed superintelligence as the next great business vibe. Well, following Elon Musk's dream, imagine widespread machine superintelligence, or something functionally close enough to it, combined with highly capable autonomous robotics. The distinction between intellectual and physical labour then collapses. The same civilisation that possesses machines capable of outperforming its best scientists, engineers and managers also possesses robotic bodies capable of manipulating the physical world.
We can already see primitive hints of the second half of that equation. Humanoid robots can now walk, run, maintain balance and perform increasingly impressive athletic movements. Videos of robots executing martial-arts-style kicks are spectacular demonstrations of control, but kicking is economically almost irrelevant. The important development comes when the same advances in balance, perception, dexterity and autonomous decision-making are turned towards work.
Then robots do not merely dance in laboratories. They enter warehouses, farms, construction sites, factories and mines. Mining is especially important to the stronger AI EAT (Economic Apocalypse Theory) scenario because an advanced machine economy requires enormous physical inputs: copper, iron, lithium, rare-earth elements, silicon, energy and the materials needed to construct data centres, power systems and millions of additional machines. If autonomous systems can explore for minerals, mine them, refine them, manufacture components, construct power infrastructure, build robots and maintain the machinery that performs those operations, automation begins closing its own material loop.
That is a much more profound development than an AI taking an accountant's job. Human civilisation presently sits between intelligence and the physical resources intelligence needs. We mine the ore, build the power stations, maintain the transmission lines, manufacture the chips, repair the machines and construct the data centres. Advanced robotics potentially removes humans from progressively more of that chain.
At the limit, superintelligent systems would not merely replace particular occupations. They could design improved machines, while robotic systems extract the materials and construct the infrastructure required to manufacture them. Machines would maintain other machines. Automated mines would feed automated refineries; automated factories would produce robots and computing equipment; automated energy systems would power them; and increasingly capable artificial intelligence would coordinate the entire network.
That is the point at which the familiar historical analogy with tractors and steam engines begins to fail. Previous machines generally automated particular human capabilities while leaving humans with comparative advantages elsewhere. The tractor replaced muscle but not the agricultural scientist. The calculator replaced arithmetic but not the accountant. The industrial robot replaced repetitive movement but not the engineer who programmed it.
General-purpose superintelligence combined with advanced robotics potentially attacks the escape route itself. If a displaced accountant retrains as a programmer, the machine may already program better. If the programmer retrains as an engineer, the machine may engineer better. If everyone is told to abandon intellectual work and become electricians, miners or plumbers, increasingly capable robotics follows them into the physical economy. Where exactly is the protected economic niche into which billions of human workers are supposed to retreat?
The standard answer is that technology has always created new jobs. That is historically true but logically insufficient. Previous technological revolutions occurred while human intelligence remained the most generally capable intelligence available and the human body remained an extraordinarily adaptable general-purpose machine. Neither condition necessarily survives the combination of superintelligence and advanced robotics.
Another objection says that human wants are infinite and therefore new employment will always emerge. Perhaps, but infinite wants do not imply infinite demand for human labour. I may want a thousand things, but if machines can supply 950 of them more cheaply and efficiently than people can, my desires have not created 950 human jobs.
Then comes the abundance objection. If machines can produce virtually everything, goods will become extraordinarily cheap. Why worry about unemployment in a world of abundance?
Because abundance and access are different things. A supermarket may contain enough food to feed everyone who walks past it, but ownership still determines who gets the food. An automated economy could theoretically produce unprecedented quantities of goods while simultaneously concentrating ownership of productive capital. The technological capacity to produce abundance does not itself establish a mechanism for distributing claims upon that abundance.
This is where Universal Basic Income (UBI) usually enters the story. If machines do the work, governments can tax their owners and distribute enough money to everyone else. Production continues, consumers retain purchasing power and humanity enters a post-work civilisation.
It is possible. But AI EAT asks a devastatingly simple question: why assume that it will happen?
UBI is not a technological consequence of artificial intelligence. It is a political decision. It requires governments to possess both the power and the continuing willingness to take a substantial share of machine-generated wealth from its owners and transfer it indefinitely to people whose labour the economy no longer requires. Nothing inside a superintelligence, robot, automated mine or automated factory guarantees that political settlement.
One should also be careful about romanticising ruling classes. History provides abundant examples of governments and elites accepting enormous human suffering when power, territory, ideology or institutional survival was at stake. Wars alone should cure us of the comforting belief that those possessing political and economic power invariably organise society according to universal human benevolence. This does not prove that future elites will abandon unemployed populations. It does show that benevolent redistribution cannot simply be inserted into the AI future as an unquestioned premise.
There is also an uncomfortable political transformation hidden inside AI EAT. Throughout industrial history ordinary people possessed bargaining power partly because elites needed them. Governments needed taxpayers and soldiers. Businesses needed workers and consumers. States needed enormous bureaucracies staffed by humans. Economic interdependence imposed constraints upon rulers.
Radical automation could weaken every one of those constraints. A population whose labour is essential possesses leverage. A population whose labour has become economically redundant possesses much less. And once advanced autonomous machines can perform not only productive work but surveillance, administration and potentially military functions, even the state's traditional dependence upon large numbers of human beings changes.
There remains an obvious problem for the darkest version of AI EAT: capitalists need customers. What is the point of owning automated factories if nobody can buy their products?
That objection has force, but it is not decisive. It assumes that maintaining mass consumer capitalism must remain the overriding objective of those controlling productive assets. An economy can change its structure. Governments can become enormous purchasers. Automated enterprises can trade with other enterprises. Owners can consume fantastically disproportionate resources themselves. Some redistribution might occur simply because preserving social stability is cheaper than suppressing disorder. None of these possibilities guarantees the continuation of the twentieth-century arrangement in which mass employment generates wages and wages generate mass consumption.
Nor does AI EAT require the extermination or deliberate starvation of billions of people. A considerably less cinematic outcome would be sufficient: permanent technological unemployment for a substantial population, falling bargaining power, increasingly concentrated ownership and a subsistence-level transfer system designed primarily to maintain political stability. That would still represent an extraordinary reversal of the economic position ordinary people acquired during industrialisation.
There are powerful reasons for caution before declaring this future inevitable. Present evidence does not show that AI is eliminating most employment, and today's robots remain far short of the general physical competence required by the full AI EAT scenario. History also demonstrates remarkable human ingenuity in inventing occupations that earlier generations could scarcely have imagined.
But neither observation refutes AI EAT because AI EAT is explicitly a scenario about what happens after those present limitations cease to hold. Pointing out that today's AI cannot replace everyone is no answer to a hypothesis whose premise is widespread superintelligence coupled to mature general-purpose robotics.
The ultimate question therefore becomes much larger than whether AI will take this year's jobs. It is whether human beings retain an indispensable economic function in a civilisation whose superior intelligence can act upon the physical world without them.
If humans remain necessary, they retain bargaining power. If machines eventually perform the thinking, mining, farming, manufacturing, construction, transportation and maintenance, and increasingly build and improve the machinery required to perform those activities, the economic relationship changes fundamentally.
The comforting answer is that such productivity will liberate humanity from drudgery and everyone will share the resulting abundance. Perhaps it will. But that conclusion requires something technology cannot manufacture: a political order determined to distribute the fruits of abundance widely.
History gives us no warrant for assuming that extraordinary concentrations of productive power will automatically produce extraordinary generosity. Superintelligence may create the technical possibility of abundance for everyone. AI EAT asks the darker question: what happens when the machines can run the civilisation, but the civilisation no longer economically needs most of the people?