AI will replace only ~5% of human work in 10 years, say Suleyman and Acemoglu
On 6 October 2026, Mustafa SuleymanDealroom has a profile for this one. Try Dealroom →, CEO of Microsoft AI, used an X Article to launch The Humanist Review, a new magazine on the future of AI published by MAI, the AI division of Microsoft. The first issue leads with an essay by Nobel laureate economist Daron Acemoglu arguing that AI will not take most jobs any time soon, and that the industry should build AI that makes workers better rather than replacing them. Benchmark's Bill GurleyDealroom has a profile for this one. Try Dealroom → shared the post, calling Suleyman "the statesman AI needs" and urging the industry to take the hyperbole "way down", arguing that doomerism and over-optimistic claims such as curing cancer in three years go hand in hand.
Key arguments from the essay:
- Slow displacement: Acemoglu estimates that only about 5% of what humans do will be replaced by AI over the next 10 years. Roughly 20% of US work (simple office and cognitive tasks) could be automated by the end of the 2020s, but only about a quarter of that is likely to be adopted within a decade.
- Modest macro impact: he expects AI to add around 1.5% to GDP over 10 years. AI is not yet visible in productivity statistics, and most adopting firms report limited gains. Meanwhile, 52% of Americans say they are worried about AI's impact on their jobs.
- History of diffusion: New York and London had power stations by 1881, yet only about half of factories and homes used electricity by the 1920s. AI adoption may be slower still because firms must reorganise around it.
- The last-mile problem: Dragon's speech recognition was near 95% accurate in 1997, yet PC dictation today is only marginally better than in 2000, after Dragon passed through Lernout & Hauspie, Nuance and Microsoft. Even 99% accuracy is not enough for full automation.
- Pro-worker AI: rather than mimicking human intelligence in pursuit of AGI, AI should complement people, for example tools that help teachers personalise lessons, which Acemoglu says are buildable with today's models.
- Policy fixes: the US taxes labour at over 25% and capital at close to zero, effectively subsidising automation. With the seven largest tech companies making up 60% of the NASDAQ, he calls for equal taxation of labour and capital, antitrust enforcement, a digital advertising tax and paying experts for their data.
The launch positions Microsoft AI's messaging around "humanist" AI, in contrast with labs that stress rapid job displacement or existential risk.
Read more: Mustafa Suleyman on X · The Humanist Review · Bill Gurley on X
