Nobel Economist: Where He Sees the Real Money in AI

Summary · 5 min read

Daron Acemoğlu is an MIT economist who won the 2024 Nobel Prize in Economic Sciences. He has done work on robots, early 20th century technology and the Industrial Revolution, and he is writing a new book on liberal democracy. In the episode he describes talking with CEOs and manufacturers about what he calls pro-worker AI.

Acemoğlu says his 2024 forecast of about 1% added US GDP from AI was built on 5% of tasks being automatable, and he would now put that share a little higher. He argues that the bottleneck is applications, not model strength, and that existing models could already support tools for nurses, teachers and electricians. He says automation is good but not enough on its own, and that raising worker productivity is a better business bet than cutting 5% of jobs. He advises students to study math, physics, economics or engineering and to build flexibility and civic responsibility. He does not expect foundation models to make trillions and sees the edge in data and applications.

Minute by minute

1. 1:11

   AI's GDP impact will arrive slowly. Asked about his 2024 forecast of about 1% added US GDP over 10 years, Acemoğlu says he would now put the share of automatable tasks a little higher than 5%. He still thinks much of the discussion overstates how fast automation will arrive. He says integrating it into organizations is hard, and diffusion of technologies that need complementary investments is slow.

2. 2:59

   Coding has clear rules, customer service does not. Acemoğlu says coding advances fastest because it happens in offices, is supervised by people fluent in technology, and has a ground truth, since code either runs or does not. He says customer service was expected to be automated first but is slow, because requests change over time, differ by person and involve gray areas. He expects humans to oversee or work with AI there.

3. 5:31

   Applications matter more than foundation models. Acemoğlu's one word for improving AI's economic effects is applications. He argues Silicon Valley puts too much emphasis on foundation models and too little on integrating applications into organizations. He says Chinese companies put engineering talent into bringing robotics and AI to production lines, while US engineering talent goes into foundation models. He also rejects the view that the current US path is the only way.

4. 9:02

   Measuring AI's impact is hard. Mogilko cites her talk with Eric Bjornoson, who found a near-zero economic impact of AI. Acemoğlu says Bjornoson is more optimistic than he is about 2034 and 2035 effects, though both expect AI to show productivity gains eventually. He agrees quality is hard to measure, but notes it was hard for cars and highways too. On well-being, he is more negative and says social media has damaged community.

5. 12:46

   Existing models can power tools for workers. Acemoğlu says current models already support useful applications, such as quality control and tools for nurses, teachers and electricians. He says the US spends almost one out of every $5 on healthcare, and that nurses could take a larger role in care if AI supplied information and quality control. He says the bottleneck is the tech industry not building these applications, not the models. Both he and Mogilko see electricians as a large market.

6. 15:09

   Government, tech firms and managers must align. Acemoğlu says healthcare and education involve government money, while tools for electricians, customer service and scientists are mostly a private-sector matter. He says CEOs respond well to pro-worker AI, but he cannot test whether they mean it because tech companies are not supplying the tools. He says tech people often answer that they are going to create AGI. He cites Foxconn, BYD and other Chinese EV makers as integrating robotics and AI, with a need for many engineers.

7. 16:56

   Niche problems beat chasing AGI. Acemoğlu agrees that opportunity lies in business problems that have not been automated, though he would not call them boring. He calls AGI pie in the sky, since nobody agrees on a definition, and says full AGI or ASI would also be dystopian. He sees an inconsistency in claiming AGI will come soon while also expecting plenty for humans to do.

8. 18:30

   Students risk delegating too much to AI. Acemoğlu uses AI for background research and for criticism of his writing, and he tells students to build an AI setup that supports their work and never replaces it. He discourages using AI to write, but encourages non-native English speakers to use it for corrections. He worries about delegation that stops people exercising their minds, and calls AI in K through 12 schooling without guardrails a dangerous social experiment. He says the calculator analogy holds only partly, because calculators took over a very small part of thinking.

9. 24:00

   Automation alone will not produce shared prosperity. Acemoğlu says he supports automation and thinks it cannot be stopped. His concern is a strategy of only automation, which would not deliver full productivity and would eliminate too many jobs without creating good new ones. He compares automating 5% of jobs and halving their cost with making workers 5% more productive, and says the latter yields more. He says successful companies in history created new markets and tasks, and that businesses that try automation often cannot make it work.

10. 28:40

    University remains worth it, with caveats. Acemoğlu thinks university will still be worth the money in 10 years, with caveats. He says not everybody should go to college, that part of what elite universities sell is a signal, and that universities must change their business model. He says high schools and middle schools are further behind, that students arrive weakly prepared, and that AI has made learning harder there.

11. 31:17

    Study foundations and build civic responsibility. Acemoğlu says human creativity needs strong foundations, so he advises studying math, physics, economics or engineering. He also advises learning flexibility, since people will need to change what they do as technology changes. He adds that schools should teach working with AI alongside civic responsibility, starting in elementary school. He says he is neither an optimist nor a pessimist and finds hope in human adaptability and shared notions of fairness.

12. 34:41

    Loss of control worries him more than existential risk. Asked about Dario's 25% chance of things going very badly, Acemoğlu welcomes lab leaders speaking up. He is more worried about the public, workers, students and parents losing control of their agency, because of how companies and labs develop and roll out AI. He says dangers have grown but so have good uses, and AI should not be decided by a handful of people. He says the media has failed worst, swinging between uncritical praise and treating AI as an enemy.

13. 40:50

    The money is in data and applications. Acemoğlu does not see how foundation models will make trillions of dollars, citing possible bottlenecks and strong open source models. He sees the edge in high-quality data and in applications that fill a niche, which are harder for open source to copy. He advises firms to capture data while compensating the workers who generate it. He adds that builders must talk to workers, since much tacit knowledge sits with them.

Published Oct 6, 2026.