The numbers come first — and in this case, the number that matters is almost zero. Despite years of AI hype, Nobel laureate Daron Acemoglu told Fortune that outside of coding, there is 'barely' measurable evidence of wide adoption: customer service employment has barely budged, and manufacturing shows the same flat line.
That inconvenient data point sits at the center of Acemoglu's frustration. The MIT economist, who has been studying AI explicitly since at least 2018, is not a denier — and he is not a true believer. He is something rarer in 2026: a man who insists on holding two ideas at once.
'I think you have to really have your head in the sand to think that AI is a stochastic parrot right now,' Acemoglu said. Frontier models are making genuine advances in comprehension, coding, and even scientific and mathematical discoveries — 'it's really great, the way proofs are being done,' he noted. At the same time, he added, 'I'm also not willing to go along with some inchoate belief that everything will work out fine.'
His new book, What Happened to Liberal Democracy?, arrives at what he calls a moment of crisis — and he draws a direct line between the dysfunction in politics and the dysfunction in the AI debate. 'We live in an environment that's been partly shaped by social media,' Acemoglu said, 'and there is a tendency to escalate everything, because that gets attention. Politics is like that. The other topic like that, unfortunately, is AI.'
On one side, he sees 'quasi-moderate-friendly' true believers so convinced AI will benefit everyone that any challenge 'drives them insane.' On the other, skeptics who dismiss the models as mere 'stochastic parrots.' Acemoglu refuses both camps. 'It's become sort of radical to hold two apparently conflicting ideas in your head at the same time,' he said.
He points to his relationship with Stanford's Erik Brynjolffson as a model for how the debate could be conducted. The two have disagreed publicly and sharply about AI's impact on productivity — Brynjolffson has argued for substantially greater gains than Acemoglu projects — yet Acemoglu says, 'Erik and I actually agree on many things.' He also credited Brynjolffson as 'the main scholar showing the potential job losses from AI,' and expressed satisfaction that his colleague has increasingly called for redirecting AI in more human-complementary ways.
Acemoglu's own research framework, developed with Pascual Restrepo, treats automation as a force that can simultaneously displace labor and create new tasks that raise worker value — an ambiguity that the loudest voices in today's debate consistently flatten into a slogan.
He also flagged what Wharton's Ethan Mollick calls the 'jagged frontier' of AI capabilities: exceptional at some tasks, unreliable at others, and requiring, in Acemoglu's words, 'a lot of detailed babysitting.'
Asked whether he is an optimist, Acemoglu declined the label. 'I wouldn't call myself an optimist,' he said. 'I would say I resolutely refuse to give up hope.'
CEO Times take: Free enterprise runs on accurate price signals — and the AI market is currently pricing on narrative, not data. When a Nobel laureate has to fight for the right to say 'it's complicated,' the deliberative infrastructure that capitalism depends on is under stress. The real risk is not that AI destroys jobs overnight; it is that a debate captured by tribal escalation produces bad policy before the technology even matures. Capital rewards clear rules. It does not reward a political culture that has forgotten how to think.
