Nobel laureate says AI will replace just 5% of jobs in a decade

Everyone is sweating over AI taking their jobs, but a newly minted Nobel laureate says you can breathe a little easier.

Economist Daron Acemoglu just published an essay in the debut issue of The Humanist Review, a new magazine from MAI. His headline takeaway: over the next decade, AI will replace only about 5% of all human tasks. The piece was amplified on X by Mustafa Suleyman.

Around 52% of Americans say they are worried about AI's impact on their livelihoods. But according to Acemoglu, that anxiety is completely out of proportion, and it is actively poisoning how the technology gets built.

The brutal reality of the last 1%

If AI were already transforming the economy, you would see it on corporate balance sheets. Right now, it simply is not in the macroeconomic productivity numbers.

Acemoglu projects that AI will add only around 1.5% to GDP over the next 10 years. That is incremental progress, not a macroeconomic earthquake.

People forget how slow foundational tech spreads. New York and London got their first central power stations in 1881, but it took until the 1920s for just half of factories and homes to wire up. AI will take even longer to change the real world, because companies have to restructure their workflows from scratch.

Then there is the reliability trap. Even if an AI agent hits 99% accuracy, full automation remains impossible. That final 1% error margin is where edge cases, legal liability, and costly human intervention live.

Trying to make neural nets mirror human cognition is a fundamental design flaw, Acemoglu argues. Machine learning and human thinking are entirely different beasts, so chasing 1:1 human imitation is burning cash on the wrong problem.

Subsidizing algorithms over workers

Acemoglu also calls out Washington and Wall Street for tilting the playing field.

In the U.S., labor is taxed at over 25%, while capital is taxed near zero. That tax wedge basically hands corporations a giant government subsidy to buy silicon and fire humans, rather than investing in staff.

Meanwhile, the top seven tech giants control roughly 60% of the NASDAQ. That unprecedented concentration of capital sucks up air and crowds out alternative ways of developing tech.

Acemoglu advocates for what he calls "pro-worker AI": software explicitly designed to augment worker capabilities rather than automate whole payrolls. And unlike full artificial general intelligence, that software can be built today.

Why it matters

For knowledge workers and everyday professionals, this essay is a sanity check. You do not need to panic about immediate mass displacement. The real day-to-day headache is dealing with messy, half-baked AI features that save ten minutes of drafting only to create twenty minutes of verification work.

For the AI industry, Acemoglu's math is a flashing dashboard light. Startups selling the fantasy of automated departments are running headfirst into the "last 1%" buzzsaw. If macro productivity numbers stay flat, CFOs are going to pull the plug on expensive AI pilots that produce clever demos instead of durable margin expansion.

In our view, Acemoglu nails the economic hypocrisy underneath the hype. The obsession with full automation looks less like a breakthrough in technical efficiency and more like a tax-advantaged financial play. Pivoting toward worker-first augmentation is not just the more ethical path; it looks like the only business model with sustainable unit economics.