By the numbers
AI and jobs, by the numbers
How much AI really affects work according to research — not the day's headline. The numbers are about task overlap, not mass unemployment. Here's what the data says, with its sources.
- How many workers does AI affect?
- Around 80% of the workforce could have at least 10% of their tasks affected by large language models; roughly 19%, at least 50%. 'Affected' doesn't mean 'eliminated': it means AI can do part of those tasks.
- How many professions are highly exposed?
- Of the 22 major occupational groups, 10 are highly exposed, 5 medium and 7 low. Knowledge jobs with lots of on-screen information lead the list; manual and in-person care work close it.
- Which sector is the most and the least exposed?
- The most exposed sector (Legal) scores 89/100; the least (Construction & trades), 8/100. A huge gap: AI doesn't reach every job equally.
- Does AI destroy jobs or change them?
- Exposure measures how much of a job overlaps with AI, not whether the job disappears. The norm isn't a profession vanishing overnight, but a change in which tasks are done and how.
- Do the most exposed lose out?
- Not necessarily. The most exposed sectors are often the ones that gain the most productivity by using AI well. What decides the outcome isn't exposure, but what you do with it.
≈80%
Source: Eloundou et al., “GPTs are GPTs” (2024)
10 of 22
Source: Our own aggregation of AIOE (Felten et al., 2021) + Eloundou et al. (2024) over 22 major occupational groups
89 vs 8
Source: AIOE index (Felten et al., 2021), aggregated to 22 occupational groups
Changes them
The opposite
Your next step
And your job? Put a number on it: the test tells you your exposure in 2 minutes, and the book what to do about it.