If AI significantly weakens the economic power of labor, democracy will need to find new ways to distribute the power that automation creates. This could mean greater taxation of profits from automation, public or collective ownership of some critical infrastructure, sovereign wealth funds, or other mechanisms through which citizens directly participate in the wealth that the automated economy creates.
By Stephen HOLMES
The debate over AI usually revolves around jobs, productivity, and fears that machines will replace humans. But an argument by Stephen Holmes in Project Syndicate moves the question to much deeper ground: what happens to democracy if economic elites no longer need the majority of citizens?
This is the central idea of what Holmes calls the “AI Curse.” It is not necessarily a scenario where millions of people are left without jobs. The risk, he says, is more subtle: that automation will weaken the dependence of capital and the state on the average citizen and, with it, the political power that this citizen has historically had. The example the author uses comes from Northern Virginia, where giant data centers for Artificial Intelligence are being built near residential communities. They consume enormous amounts of energy, while local communities face higher costs and pressure on infrastructure. The question Holmes raises is not simply whether a data center is good or bad for the local economy. It is much more political: do the owners of these structures still need the people who live around them?
This question is related to the way modern democracy is constructed. Historically, economic and political power has not been forced to tolerate citizens’ rights solely out of altruism. In many cases, it has been dependence on citizens that has produced compromise. The state needed people to pay taxes, work, serve in the military, and produce. Industry needed workers not only as labor, but also as consumers.
This created a reciprocal relationship: the citizen gave something to the state or economy and, in return, demanded rights, wages, public services, and representation.
Even many of the social reforms that seem self-evident today can be seen through this logic. Workers organized because industry needed them. Armies needed millions of citizens, and this made it more difficult for those citizens to be treated simply as subjects. Even Henry Ford’s model, with its high wages for the time, was not simply an act of humanity: industry needed a broad mass of consumers to buy the products it produced. So economic dependence also produced a kind of political ‘leverage’. AI can change this balance.
If a company can produce more and more with less and less human labor, then the bargaining power of the worker changes. A strike is effective when the employer needs the worker. A threat to not pay taxes has weight when the state needs the tax base. Even the consumer has a kind of power when the economy depends on millions of consumers.
But what happens in an economy where most of the value is created by capital, algorithms, robots, AI models, and infrastructure controlled by a relatively small number of companies? Herein lies Holmes’s most interesting argument: AI doesn’t need to make humans redundant to weaken democracy. It just needs to make the owners of capital less dependent on them. That’s a very important distinction.
A society can be extremely wealthy and at the same time politically less democratic. If automation creates so much wealth that the majority of the population can be supported through transfers, subsidies, or some kind of social “reimbursement,” consumption can continue even without mass employment. But a citizen who receives a payment from someone else’s wealth has a very different bargaining position than a citizen who works, pays taxes, organizes, and can withdraw his or her labor from the economy.
Here Holmes connects AI to a well-known concept in political economy: the resource curse. In countries with oil, gas, or other large natural resources, governments may have less need to tax citizens. And when a country doesn’t need to collect much in taxes from its population, one of the classic relationships that has produced political representation weakens: “no taxation without representation.” In AI’s version, the resource is not oil. It’s chips, models, data, energy, data centers, and capital. This doesn’t mean that the United States is turning into a petro-state. The analogy has its limitations. But the mechanism Holmes highlights is valid: the more wealth that can be created from assets that require less direct participation from the population, the more we need to think about how citizens will maintain their bargaining power.
And that’s where data center politics comes in. A data center may not have thousands of permanent workers, but it does require land, energy, water, a power grid, and local permits. That means communities still have one card in their hand: they can say no. Resistance to data centers isn’t just a revolt against technology. It may be one of the last forms of bargaining power communities have in the face of an industry that has enormous capital but relatively little need for local workers.
The problem is that even this power can be temporary. If federal or state governments decide that AI infrastructure is of strategic importance, they can limit the powers of local governments, use other mechanisms to accelerate the construction of projects. At that point, the community that had the right to say “no” could lose that power as well.
And here the argument becomes truly political. If AI brings about a dramatic increase in productivity, the question is not just who will work, but who will own the profits of that productivity. If the profits are concentrated in the hands of the owners of the models, chips, data centers, and capital, while the rest of society is compensated largely through transfers, then a new political relationship could emerge: citizens are no longer co-producers of wealth, but beneficiaries of it. This is the shift that must be taken seriously. This scenario is not necessarily going to happen. AI could create new professions, increase worker productivity, and generate new industries that do not exist today. Even historically, new technologies have eliminated professions and created others.
But even if AI doesn’t bring mass unemployment, the concentration of ownership could be a bigger problem than automation itself. Ultimately, the question isn’t whether machines will replace humans. The question is whether humans will continue to have something that others need to take from them.
Labor has been one of the main sources of this power for centuries. Taxes, collective organization, voting, and consumption have been others. If AI weakens the economic power of labor in particular, democracy will need to find new ways to distribute the power that automation creates. This could mean greater taxation of profits from automation, public or collective ownership of some critical infrastructure, sovereign wealth funds, or other mechanisms through which citizens directly participate in the wealth that the automated economy creates. Because the biggest threat from AI may not be a bot taking jobs. It may be an economy so productive that it no longer requires the economic participation of most citizens.
And if that happens, the question that will define the future of democracy will not be “will AI replace us?” It will be much more difficult: If the owners of the machines no longer need our jobs, what will compel them to still need our citizens?