How Is AI Affecting College Major and Course Choices? Evidence from Finland (IZA DP 19056)
A · Strong evidence Aalto and Helsinki economists use Finnish administrative data on university applications, course choices and graduate employment with event studies and several exposure measures; method and robustness checks are shown.
Summary based on the full text·Original: about 55 minutes to read
In short
IZA Discussion Paper 19056 (October 2026) by Ciprian Domnisoru (Aalto) and Tuomas Heikkinen (Helsinki). After ChatGPT 3.5 launched in November 2022, Finnish applications to AI-exposed fields fell and kept falling, most sharply in computer science and other ICT. Non-ICT students took fewer programming and algorithms electives and more AI/ML electives. The application drop is much larger than the rise in ICT graduate unemployment, which points to expectations rather than current conditions.
Key numbers
- 38.0 per centfall in first-choice applicants to ICT programmes, 2022 to 2026 rounds, while other fields rose 16.5 per cent
- 2.7 to 5.6 per centunemployment of ICT master's graduates under 30 after 2022
- one percentage pointlower employment rate per standard deviation of AI exposure for graduates of exposed majors, by 2026
- 14.02 per centdecline in first-choice applicants per standard deviation of AI exposure in 2026
As reported by the source.
English·Web page·opens iza.org
What it reports
- First-choice applicants to ICT fell 38.0 per cent between the 2022 and 2026 rounds; other fields rose 16.5 per cent, business, administration and law 39.8 per cent. ICT seat supply was roughly unchanged.
- Event-study declines for ICT relative to other programmes: -34.36 per cent in 2025, -45.34 per cent in 2026.
- Across all degrees, one standard deviation more AI exposure goes with 8.79 per cent fewer first-choice applicants in 2025 and 14.02 per cent fewer in 2026; excluding ICT barely changes this.
- ICT programmes went from about 2:1 oversubscribed in 2019 to close to 1 in 2026; one university left 112 computer science seats unfilled in 2026.
- Unemployment of master's graduates under 30 rose in every field after 2022: ICT from 2.7 to 5.6 per cent, engineering from 2.1 to 6.4. ICT remains mid-ranked; arts and humanities sits at 14.2 per cent.
- By 2026, one standard deviation more exposure goes with one point lower employment and about 0.7 points higher unemployment for graduates of exposed majors. With firm-by-month fixed effects the gap disappears, so the authors attribute it to hiring across firms, not reallocation within them.
- Non-ICT students reduced programming electives, especially those who skipped mathematics in the high school exit exam; AI/ML elective take-up rose from 0.3 to 0.7 per cent, more among men.
How they got it
Administrative records on university applications, course enrolment and employment, analysed with event studies around November 2022. Exposure uses Handa et al. (2025), with checks on Eloundou et al., Frey and Osborne and a Finland-specific automation measure, plus controls for remote-work exposure and field unemployment.
Limits
- ICT treatment covers four degree codes in one broad field, so standard errors rest on few treated clusters; the authors add randomisation inference.
- Other exposure measures give effects about half the size or less.
- Expectations are inferred from the gap between applications and labour outcomes; the paper says the response "may be driven" by them, it does not measure expectations directly.
- Applications fell but enrolment did not, since most programmes still receive more applicants than seats.
Why it matters for companies
The pipeline for technical hires is shifting now, years before graduate labour markets show much damage. Employers that need software and data skills in 2028-2030 should expect fewer applicants from domestic programmes, and junior candidates with fewer programming electives and more AI coursework. The employment effect runs through hiring that does not happen at the firm level, which internal workforce data rarely shows.
Summary based on the full text. Read the original for the full argument and data.Read the original on iza.org
- Tags
- Future of work, Education