AI, Work, and Power: The Promise No One Voted For

A tech executive announced that within eighteen months all office work could be automated. This wasn't said by an apocalyptic blogger or a nervous union organizer. It was said by the head of artificial intelligence at one of the largest companies on the planet. The interview circulated in financial media as if it were just another piece of news, sandwiched between quarterly earnings and stock quotes.

That's where the first contradiction appears. The very people building these tools announce, with a certain enthusiasm, the disappearance of the work they themselves were doing not long ago. Sam Altman wrote that we're entering "the gentle singularity," a moment when artificial intelligence becomes so capable that human labor gets completely reorganized. Yann LeCun later pushed back on that optimism. There isn't even consensus among the people building the technology. That alone says quite a bit.

What does generate agreement are the numbers starting to appear. A Stanford University study identified measurable effects on employment, concentrated among young workers in occupations exposed to automation. This isn't a prediction. It's a finding about what's already happening.

How new is this moment, really? This isn't the first time a technology has promised to free us from grueling labor only to end up redistributing power in ways nobody voted for. Economic historian Carl Benedikt Frey documented every wave of automation, from mechanical looms to industrial robotics. Those productivity gains concentrated among those who owned the capital. Workers absorbed the social adjustment: unemployment, forced migration, falling wages. It was never automatic that technology would benefit everyone. It was a political decision, made over and over again, not to share those gains.

These patterns repeat in other contexts too. Today's rhetoric reproduces, almost word for word, the technological optimism of earlier eras. Aldous Huxley warned about a different kind of domination: not one that represses through fear but one that distracts through pleasure and comfort until you stop asking who benefits. Neil Postman picked up that idea. Today we might add technological convenience as the most efficient way to anesthetize critical thinking. No one needs to censor anybody if people are too busy adapting to the next tool.

The infrastructure is literal. The International Energy Agency detailed how much electricity the data centers training these models consume. The figures are significant enough to reframe entire conversations about energy policy. The Electric Power Research Institute reached similar conclusions: demand is growing faster than generation capacity in several regions. A handful of companies are redirecting resources once earmarked for hospitals, schools, or housing. They're directing them toward servers whose clearest measurable effect, so far, is reducing the need to hire people.

And who really benefits? Bloomberg Intelligence surveyed technology directors at ninety-two banks. AI-driven automation is becoming the primary route to cutting headcount in the financial sector. Standard Chartered reported it openly in its earnings statement. That term "operational efficiency" is a transparent euphemism for fewer employees. You just have to read the financial reports out loud.

The rhetoric of efficiency and optimization tends to come paired with a redistribution of value toward those who already held the capital. The difference this time lies in the speed and the scale. A model gets updated every few months and replaces analysis, writing, programming, and customer service tasks across dozens of industries simultaneously. This is more complicated than it looks.

Daron Acemoglu's argument is useful because it avoids both blind optimism and doom-mongering. The economic growth this technology promises is real, though more modest than its promoters announce. Without deliberate redistribution policies, that growth will concentrate even further. It's an uncomfortable diagnosis. This is going to happen. The question is who decides how it gets divided up.

That same question intersects with what the Marienthal experiment reveals — the Austrian town where researchers studied the psychological effects of mass unemployment in the 1930s. Income is necessary for survival. But it alone isn't enough to sustain the sense of purpose, social structure, and dignity that work provides when it functions well. The gap between what's necessary and what's sufficient is where the real debate over automation is being fought. Almost no one holds that debate with that kind of honesty.

Pope Leo XIV touched on a point in an encyclical that usually gets left out of financial reports: technology isn't neutral just because it's efficient. Its moral value depends on whether it expands human dignity or simply redistributes power toward those who already hold it. It's an old argument, almost as old as Juvenal's satire on bread and circuses in imperial Rome. Keep people distracted and fed just enough and no one will ask who's actually in charge.

I don't have a clear answer for how to solve this. The solution never came from waiting for the market to regulate itself. It came from communities and governments that actively decided to build redistribution mechanisms before the crisis became irreversible. That's not nostalgia. It's what the historical record shows again and again, whenever someone bothers to look at it carefully.

Stones don't lie, but historians sometimes do.

Will we decide collectively, and in time, who gets to keep the gains from this automation?

1. Brynjolfsson, Erik, Bharat Chandar, and Ruyu Chen. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, November 2025.

2. Acemoglu, Daron. The Simple Macroeconomics of AI. NBER Working Paper 32487, April 2024.

3. Frey, Carl Benedikt. The Technology Trap: Capital, Labor, and Power in the Age of Automation. Princeton University Press, 2019.

4. International Energy Agency. Energy and AI. Paris: IEA, April 2025 (updated 2026).

5. Bloomberg Intelligence. Survey of technology directors at ninety-two banks, January 2025.