What if the greatest political danger of artificial intelligence is not that machines become more powerful than humans, but that humans become increasingly unwilling to share the wealth those machines create?
That question runs through a new mathematical-economic model developed by Nobel laureate in economics Daron Acemoğlu, together with Arda Gitmez and Mehdi Shadmehr. Their research examines what happens when human labour is progressively displaced by machines, capital becomes more concentrated and inequality between workers and capital owners increases. Its conclusion is deliberately provocative: “Our main result is a complementarity between automation and repression.”
The paper begins with a familiar economic division. Entrepreneurs own capital — factories, robots and computer programmes — while employees supply labour and receive wages. As automation expands, machines take over more tasks. But the economic consequences extend beyond productivity. If capital income grows significantly faster than wage income, inequality rises and, with it, the possibility of revolt against existing property relations.
That creates a fundamental political choice for what the paper calls the capitalist state. It can redistribute wealth to workers through taxation and social welfare, reducing their incentive to revolt. Or it can repress political action, reducing the ability of those who have lost economic power to challenge the existing system.
The distinction becomes increasingly important as automation advances. More automation can mean more workers losing jobs or accepting lower wages. To maintain political stability through redistribution, the state must therefore transfer an increasing share of the gains from automation. From the perspective of capital owners, redistribution functions as an additional cost. Repression offers a different calculation: suppress political resistance rather than continually paying to offset the unequalising effects of automation.
The paper identifies a feedback loop that makes the argument more disturbing. More capital makes automation more profitable because additional capital can perform more tasks. Greater capital accumulation therefore encourages greater automation. Greater automation increases the capital share of national income and can intensify inequality. Greater inequality increases the likelihood of revolt. The resulting political threat makes repression more attractive — and repression then creates greater room for further automation.
As the authors put it: “A higher capital stock in turn encourages more automation and thus more repression.”
This is not simply a model of technological unemployment. It is a model of how economic incentives can alter the political choices surrounding technological change. The paper argues that decentralised firms do not take into account the political risks created by their individual automation decisions. They can therefore “over-automate and underinvest in new tasks”. The market may produce more automation than is politically sustainable because individual capitalists do not fully account for the consequences for inequality and the probability of revolt.
The state, acting on behalf of capital owners, faces a different calculation. The model finds that it can regulate automation, redistribute income or repress political action. Without repression, the capitalist state chooses an automation level below the output-maximising level because excessive automation increases political risk. With repression, however, the state can adopt an automation level above the output-maximising one because it no longer needs to account for changes in the risk of revolt in the same way.
That produces one of the paper’s central findings: “Repression leads to the capitalist state adopting a level of automation above the output-maximizing one.”
Redistribution remains an alternative, but its cost rises as automation expands. The paper finds that a greater capital stock raises both the level of automation and the level of redistribution required to prevent revolt. At a certain point, capital accumulation and automation create a relative advantage for repression compared with redistribution.
The long-run implications are particularly striking. The economy can begin under a redistribution regime and remain there for a period. But as capital continues to accumulate and automation increases, the model allows for a transition towards repression once a critical level is reached. Alternatively, if the threat of revolt is sufficiently strong from the beginning, repression may be chosen immediately.
The paper summarises the trajectory bluntly: “In the long run the economy tends to repression.” It also states that “Society converges to repression”, either because repression is chosen from the beginning or because redistribution is eventually abandoned once the capital stock reaches a critical level.
The most controversial extension concerns democracy. The model suggests that increasing capital accumulation can make democratic decision-making progressively less acceptable to capital owners because the burden of redistribution becomes greater. Under specified conditions, capital owners may eventually prefer to overthrow a democratic government.
“Capital accumulation and thus greater automation encourages the capitalists to support a coup against democracy and set up a repressive system,” the paper states.
That argument inevitably recalls Karl Marx. In the Communist Manifesto, published in 1848, Marx described the state as “the committee that administers the common affairs of the whole bourgeoisie”. His broader argument treated the state as closely connected to the interests of the dominant economic class. The new research does not establish Marx’s thesis as a historical law. Instead, it constructs a mathematical mechanism through which growing economic power could create incentives for political repression.
The authors are explicit about the limits of their model. It assumes that automation weakens the economic power of workers, a proposition that is not necessarily valid in every circumstance. It also assumes that capital owners are motivated by profit and can effectively control the state apparatus. Real political systems are more complicated. The model therefore cannot mathematically prove that democracy will gradually erode, and the authors do not claim that it does.
Yet the model becomes more provocative when artificial intelligence enters the discussion. The paper argues that technological change and automation, especially with artificial intelligence, can make repression more cost-effective. If that mechanism holds, the same technologies that increase productive capacity could also strengthen the economic incentives for political control.
The paper even tests whether the creation of new labour-intensive tasks changes its central conclusion. It does not. The authors find that firms underinvest in new tasks relative to what the capitalist state would choose, but the fundamental complementarity between automation and repression remains.
The paper’s final message is therefore less a prediction than a question requiring evidence. Its authors call for systematic empirical research into how automation and technological change affect politics through job losses, inequality, protests, revolt and collective action.
For Marx, the struggle surrounding machinery was already a political struggle over the means of production. More than a century later, Acemoğlu and his co-authors offer a mathematical model in which that old conflict acquires a new technological dimension. Their most unsettling proposition is that as capital becomes more powerful and automation more pervasive, the political cost of redistribution may rise at precisely the moment when redistribution becomes most necessary.
And if redistribution becomes too expensive, the model asks whether the state will choose the alternative.
“More automation requires more redistribution in order to counterbalance the unequalizing effects of automation. Repression avoids this additional cost of automation.”
That is the paper’s starkest warning: the machines may not need to overthrow democracy. The economic system built around them could, under certain conditions, create incentives for people with the most to gain from automation to do it themselves.
