Does More Schooling Mean More AI Exposure? A 177-Country Look

Why schooling and AI exposure are linked

Our AI exposure score runs from 0 to 10 and says how much of a country's work could be done or helped by AI. We weight each occupation group's score by the number of workers in it. Work that mostly involves words, numbers and screens, such as professional, technical and clerical jobs, scores high. Work with hands and bodies, such as farming and building, scores low.

The jobs that score high usually require more years in school. A country whose adults average 13 or 14 years of schooling is a country with many people in professional and office jobs. So we would expect exposure to rise with schooling. The data agree. Below we look at how strong the link is, where it breaks, and what it may say about the next generation.

Adults' schooling and exposure

The dots rise from left to right. At the left are countries such as Niger (1.4 years), Mali and Ethiopia with exposure scores near 3. At the right are Germany, Switzerland, the United States and Iceland (about 14 years) with scores of 5 or more. Each extra year of adult schooling goes along with about 0.17 more points of exposure, on average.

The two right-hand columns explain why. In countries where adults average 12 or more years of schooling, about 53 of every 100 workers do office-type jobs (managers, professionals, technicians, clerks) and about 3 farm. Where they average under 4 years, about 12 in 100 do office-type jobs and about 43 farm. Job mix, not school alone, is the direct cause of the score.

Literacy: a step rather than a slope

Adult literacy tells a different story from years of schooling. Below 90% literacy the average score barely moves, from 3.40 to 3.50. Above 90% it jumps to 4.19, then 4.37 at 99% and above. The correlation is 0.56, weaker than for years of schooling. One reading is that basic literacy is a threshold: once nearly everyone can read, a country's job mix can start shifting toward the work AI touches, while further years of schooling push it further.

The next generation: schooling catch-up

The Human Development Report also publishes expected years of schooling: how many years a child starting school today can expect to get if current patterns continue. In the median country that is 13.2 years, against 9.7 years of mean schooling for adults today, a difference of 4.3 years.

In several big countries, today's children are expected to get 6 to 8 more years of schooling than their parents' generation. In China, adults average 8.0 years and children can expect 15.5. In India it is 6.9 and 13.0. This does not say what jobs those children will find. It does say that, if a more educated generation moves into jobs that match their schooling, the job mix in these countries has a lot of room to move toward the kinds of work our scores rate higher. That is a what-if, not a forecast, and it depends on whether enough such jobs exist.

Countries that break the pattern

As in our other analyses we only use countries that report all nine major occupation groups here, so a missing group cannot distort the average.

More exposed than their schooling predicts

Less exposed than their schooling predicts

The post-Soviet countries are a good example of the second group: Tajikistan, Moldova and Kyrgyzstan have high schooling (11 to 12 years on average) but job mixes with a lot of manual work: 33% of Tajikistan's workers are in elementary jobs, 50% of Moldova's farm, and 17% of Kyrgyzstan's are in crafts and trades. Their scores are lower than schooling alone suggests. Schooling opens the door to office work only if office jobs exist.

What this means

For individual workers, a degree does not mean a job is "at risk" or "safe." Most exposed occupations will probably change, not vanish, and more educated workers are often better placed to use the tools. For countries, the chart shows a pattern worth planning around: as schooling rises, the nature of work shifts toward tasks AI can already help with. The key question for ministries of education is not whether to educate more people, but whether schooling prepares them to work alongside these tools, and whether the economy creates enough skilled jobs to absorb them.

Limits to keep in mind

• Correlation is not cause. Schooling does not itself raise exposure. Both reflect the type of economy.
• Data years differ. Schooling figures are for 2023. Literacy years vary by country (World Bank, 164 countries with data). Employment years vary.
• Scores are model-estimated. They say how much of a job could be done with AI, not how much will be.
• Schooling years are not quality. A year of school in one country is not the same as in another.

Spotted an error or know a better source? Report it and we will check it against the source data.