Microsoft shares Acemoglu’s estimate: AI adds 1.5% to GDP

The bottleneck is work, not model size

Acemoglu argues that AI that complements people’s skills will raise productivity more than full automation. Even 99% accuracy may fall short once systems meet the final stretch of deployment and the actual needs of users.

He says the organizational transition could take longer than the shift to electricity. Bigger models, in his view, will not solve that problem. Companies need applications that are easy to deploy and change how production works.

A forecast that suits Microsoft

The piece appeared on Microsoft’s corporate blog, under the unusual title The Humanist Review of AI. Contributors sign their articles by hand.

The estimate also fits a strategy built around adding AI to existing products, rather than betting on automation at scale. I think the more consequential point is the gap between technical performance and workplace productivity: a system can be highly accurate and still fail to change how a company operates.

What the post leaves open is how much of the projected gain depends on applications that genuinely reorganize work, rather than simply adding AI features to existing tools. If adoption takes longer than the transition to electricity, the 1.5% forecast may be less a verdict on AI’s capabilities than on the time and effort required to put them to use.