The Pulse
Microsoft Publishes Acemoglu’s 1.5% AI GDP Forecast
Microsoft’s Humanist Review of AI published Nobel economist Daron Acemoglu’s estimate that AI could raise GDP by about 1.5% over 10 years. Acemoglu argues that slow workplace adoption and the difficulty of reorganizing jobs will limit near-
AI.info Team · Written with AI ·
AI could raise GDP by about 1.5% over 10 years, according to Nobel economist Daron Acemoglu—a restrained estimate published by Microsoft’s Humanist Review of AI on July 15. The figure describes a projected increase in GDP over the decade, not an annual growth rate. The Decoder highlighted the essay on October 6, framing it as a notably cautious forecast amid larger claims about AI’s economic payoff.
Acemoglu’s 5% estimate starts with tasks
Acemoglu’s forecast pairs that GDP estimate with a projection that AI would replace about 5% of human work within 10 years. He calls the figure a “guesstimate,” based on how AI capabilities might develop and how quickly firms adopt new technology. His calculation begins with a forecast that simple office and cognitive tasks could make up about 20% of work in the U.S. economy by the end of the 2020s.
He then estimates that only about a quarter of those automatable tasks would be automated within a decade, producing the 5% figure. The distinction matters: his estimate is about the share of work tasks that might be automated, not a prediction that 5% of workers will lose their jobs. Acemoglu also cautions that AI progress has moved faster than he expected, while saying that widely usable applications capable of changing production remain scarce.
Why better models may not lift output quickly
The essay’s central argument is that technical capability alone does not translate into economy-wide productivity. Companies must decide how to divide tasks between workers and AI, train employees to use new tools, and change internal processes. Acemoglu argues that those organizational changes can take years, particularly when a technology demands a fundamental redesign of work.
He points to electricity as a precedent: factories and homes did not adopt it widely as soon as generating stations appeared. AI adoption may take longer still, he writes, because firms must work out where the tools fit and how managers should use them. The essay says AI’s gains have yet to appear clearly in productivity statistics, while noting that results in controlled settings can look better than performance in ordinary workplaces.
Automation is not the same as assistance
Acemoglu argues that AI developers and companies are focusing too heavily on replacing workers. He favors “pro-worker AI”: systems that help people handle more complex tasks or make better decisions rather than trying to take over entire occupations. Such tools, he says, could expand what workers can do and create productivity gains without requiring full automation.
The essay uses education to make the distinction concrete. Rather than trying to replace teachers, an AI tool could identify which material is giving students trouble and help teachers organize targeted support. Acemoglu says his research team has built prototypes of this approach and is evaluating them. He presents that kind of focused application as more feasible than automating teaching outright.
A forecast with a stated caveat
Acemoglu does not present the 1.5% estimate as a precise prediction. He says it depends on assumptions about the pace of AI development and adoption, both difficult to forecast, and acknowledges that model advances have been faster than he expected. His broader point is that the economic effect depends not only on what models can do, but on whether firms build applications that fit real work.
“My forecast (with the full warning that it was and remains impossible to know how quickly these models would advance) was that GDP would increase by about 1.5% over 10 years.”
Daron Acemoglu, economist and MIT Institute Professor
The forecast also separates the potential of AI from the business choices shaping its use. Acemoglu argues that tools designed to complement workers could have a larger effect than systems aimed at automating whole jobs. His classroom example is specific: an AI tool identifies students’ learning gaps so teachers can direct support where it is needed.