When people talk about who will adapt to AI at work, education usually comes up first. The assumption is simple: people with more schooling will understand the technology, trust it more and use it better. I wanted to check that assumption against real data, so I spent part of this year on an independent analysis of European survey data.
The data
I used Special Eurobarometer 554, a survey the European Commission ran in spring 2024 with 26,404 respondents across all 27 EU member states. It includes a module on artificial intelligence and the future of work: how people feel about AI in the workplace, how they rate their own digital skills, and whether they have already worked with AI-supported tools. I fitted survey-weighted regression models with country fixed effects, so the comparisons happen within countries rather than between them.
Three things stood out
Digital competence matters more than education. How confident people feel with digital tools was the strongest predictor of positive attitudes toward AI at work, well ahead of education or age. A diploma, on its own, says surprisingly little about who feels ready.
Experience builds acceptance. People who had already met AI at work were more accepting, not less. Familiarity seems to reduce fear more than it confirms it.
Precarity works through its own channel. People in financially precarious situations were less accepting of AI at work, and digital competence did not explain that gap. Feeling economically exposed shapes how people see the technology, whatever their skills.
These are cross-sectional data, so the results describe associations, not causes. Digital competence is also self-rated, which means it partly measures confidence. That is interesting in itself, but it is not the same as a skills test.
Why this matters for AI learning
If feeling competent matters more than formal education, then AI readiness is something people can learn. That is good news for anyone who designs courses, workshops or tools. It moves AI literacy from a fixed trait of the educated to a skill that can be taught.
The precarity result is the warning. Training alone will not close the gap for people who feel their jobs are at risk. An AI literacy programme that ignores economic insecurity will mostly help the people who were already comfortable.
This is the question I keep coming back to in my research: not whether AI helps people learn, but who it helps and who it leaves behind.
The analysis code is on GitHub, and the working paper is in progress.