⚖️ Not Poor. Just Irrelevant.

How AI Makes the Middle Class Not Poor, but Irrelevant

Anyone writing about AI and robotics almost inevitably ends up at extremes. One extreme is utopian: machines take over work, people live free and fulfilled, material prosperity for all. The other is dystopian: mass unemployment, collapsed purchasing power, societal destabilization. Both scenarios are technically plausible. But both are of limited use as endpoints, because the most likely outcome lies in between. And the in-between is the most dangerous, because it triggers no clear warning.

I want to develop a framework in six theses that takes all three possibilities seriously. Not as a prediction. As a thinking aid for a question I consider the most important of the coming years: not what AI can do, but who decides how it is deployed.

Thesis 1: There Are Two Endpoints and a Middle Path. The Middle Path Is Not an Equilibrium.

I call the first scenario Rome Without Slaves. The Roman Empire at its peak functioned through slave labor: a small elite lived in prosperity because a large mass did the work. What AI and robotics enable is structurally the same model, just without the moral price. If machines produce, transport, care, and administer, all people could live in conditions previously reserved for a privileged minority. Bread and circuses, but for everyone. This is not fantasy. It is a real technical possibility.

Citrini Research has described the second scenario as a Global Intelligence Crisis. AI increases productivity and corporate profits but destroys purchasing power faster than it creates prosperity. Growth that shows up in statistics but never reaches people's lives. The authors call this Ghost GDP. Those who don't own machines no longer own anything. The system devours itself.

In between lies the most likely scenario: a divided society. Not everyone ends up in paradise, not everyone in collapse. Those who understand the technology, deploy it, and help shape it live in the first scenario. Those who remain passive experience the second. This division is not a linear transition. It is a tipping point. And tipping points rarely announce themselves loudly.

»The most likely path is neither utopia nor dystopia. It is a division. And divisions rarely announce themselves loudly.«

Thesis 2: Classical Capitalism Is Poorly Equipped for All Three Scenarios. This Is the Real Structural Problem.

Classical capitalism rests on a cycle: companies pay wages, workers buy products, profits are reinvested. What happens when a growing share of production is taken over by systems that receive no wages and buy nothing? This is not an ideological question. It is an accounting one.

The decoupling is already measurable, long before AI has reached its full potential. Acemoglu and Restrepo demonstrated in a widely cited study in Econometrica that 50 to 70 percent of the increase in wage inequality in the US between 1980 and 2016 is attributable to automation. The productivity gains were minimal. The ILO estimates the global wage gap for 2024 at 2.4 trillion dollars annually, the amount that would have additionally flowed to workers worldwide had the labor share maintained its 2004 level. The labor share fell during this period, it did not rise. AI accelerates this trend; it does not create it.

David Autor of MIT describes the result as Hollowing Out: the labor market loses its middle. Not through mass unemployment, but through the displacement of mid-level qualifications into poorly paid service work. People don't become poor. They become economically marginal. The IMF showed in a 2024 simulation study that AI adoption intensifies inequality in all modeled scenarios, because capital returns rise disproportionately. There is no scenario in which this effect resolves on its own.

There is a historical precedent for this condition. During industrialization between 1790 and 1840, productivity in England rose significantly while workers' real wages stagnated. Economic historians call this phase the Engels Pause. It ended not through market mechanisms but through political reforms, union formation, and state social policy. That took half a century. The question is whether we can or want to afford a similarly long pause.

Thesis 3: Orwell and Huxley Are Not the Past. They Are Two Possible Architectures of the Future.

George Orwell's 1984 and Aldous Huxley's Brave New World are usually read as historical warnings. I read them as technical blueprints that become fully realizable for the first time through AI.

The Orwell model controls through surveillance and pain. AI makes this model scalable like never before: facial recognition, communications analysis, social credit systems that reward compliance and punish deviation. What authoritarian states practice systematically today appears in democratic societies as creeping normalization: work tracking, algorithmic decision systems, surveillance disguised as convenience.

The Huxley model controls through satisfaction. No oppression, no pain. Instead, a world in which all needs are met, distraction is abundant, and no one demands political participation anymore because desires are fulfilled. AI and robotics can deliver this model. It requires no dictator. It only requires a world in which nothing is dissatisfying enough to motivate resistance.

The division scenario mixes both models. A small group lives in Huxley mode, comfortable and distracted. A larger group experiences creeping Orwell elements: algorithmic control, performance monitoring, restricted options, without anyone labeling this condition as oppression.

»The division scenario mixes both models. Huxley at the top, creeping Orwell at the bottom. Without anyone calling it that.«

Thesis 4: Democracies Are Not Built for This Speed. This Is Not Criticism. It Is a Diagnosis.

Democracies decide slowly. This is not a weakness but by design. Slow decisions protect minorities, force consensus-building, prevent rapid mistakes. In times of gradual change, this is an advantage. When the economic foundation of a society shifts within a few years, it becomes a vulnerability.

The argument of the benevolent dictator, a system architect who makes the right decisions quickly and without veto players, is not new. It fails historically on a simple fact: benevolence is not a property of institutions; it is a property of individuals. Individuals die, become corrupted, or make mistakes. What remains is the instrument without the intention.

The division scenario arises not from bad intentions but from institutional inaction. Parliaments debate while investment decisions are being made. Regulation follows technology with years of delay. This is not the failure of individual politicians. It is a structural property of systems designed for consensus in a world that no longer expects consensus.

Thesis 5: The Real Power Question Is Not Democracy Versus Autocracy. It Is: Who Owns the Infrastructure?

There is a shift in the power architecture of modern societies that has not yet arrived in the political debate. Traditionally, power was tied to territory. States controlled their territory, and whoever controlled the territory controlled the economy. This equation no longer holds.

The companies that operate AI infrastructure, own computing capacity, and develop models exercise power that is not territorially bounded. A company operating the AI systems on which an economy depends holds a position of power that no voting right can relativize. If Europe primarily sources AI infrastructure from American or Chinese providers, European digital sovereignty is a claim, not a reality. This question is rarely posed so directly because it has uncomfortable answers.

»Whoever owns the infrastructure owns the basis for decision-making. This is the fundamental political question of the next decade.«

Thesis 6: The Window Is Still Open. But It Requires Us to Ask the Right Question.

I am not a pessimist on this matter. We find ourselves in a moment where the decisive course-settings are still pending. That is the difference from scenarios that have already become self-fulfilling.

But I consider it a mistake to treat the utopia scenario as a likely outcome without naming the conditions under which it can occur. Three of these I consider non-negotiable. First, an honest debate about ownership of automated value creation. A system in which productivity gains do not circulate will collapse, regardless of its political label. Second, an infrastructure strategy that understands sovereignty as an investment decision, not as rhetoric. Third, societal institutions that conceive of work beyond the wage relationship. Those who do not answer this question leave the answer to the market. And the market does not optimize for societal stability.

The division I describe as the most likely middle path is not a law of nature. It is the result of decisions that are either made or not made. Those who shape will land in the first scenario. Those who wait will land in the third.

Two extremes, a likely middle path, and a window. It is closing.

Sources

1. Citrini Research / Alap Shah: The 2028 Global Intelligence Crisis. February 2026. citriniresearch.com
2. Orwell, G.: Nineteen Eighty-Four. Secker & Warburg, 1949.
3. Huxley, A.: Brave New World. Chatto & Windus, 1932.
4. Acemoglu, D. / Restrepo, P.: Tasks, Automation, and the Rise in U.S. Wage Inequality. Econometrica 90 (5), 2022, pp. 1973–2016.
5. Autor, D.: Applying AI to Rebuild Middle Class Jobs. NBER Working Paper 32140, 2024. nber.org
6. Cazzaniga, M. et al. (IMF): Gen-AI: Artificial Intelligence and the Future of Work. IMF Staff Discussion Note SDN/2024/001, January 2024. imf.org
7. Moll, B. / Rachel, L. / Restrepo, P.: Uneven Growth: Automation's Impact on Income and Wealth Inequality. Econometrica 90 (6), 2022, pp. 2645–2683.
8. Karabarbounis, L.: Perspectives on the Labor Share. Journal of Economic Perspectives 38 (2), 2024, pp. 107–136.
9. ILO: World Employment and Social Outlook: September 2024 Update. Geneva: ILO, 2024. ilo.org
10. Özkiziltan, D.: Governing Engels' Pause: AI and the World of Work in Germany. ILR Review 77 (5), 2024, pp. 846–856.
11. Acemoglu, D. / Johnson, S.: Power and Progress. PublicAffairs, 2023.
12. Zuboff, S.: The Age of Surveillance Capitalism. PublicAffairs, 2019.
13. Keynes, J.M.: Economic Possibilities for our Grandchildren. 1930. In: Essays in Persuasion, 1931.