Unemployment insurance for the AI transition
Unemployment insurance and job search appear on 86 recent papers, but only 20 of them carry an AI, automation or algorithm keyword. If language models displace office workers, the benefit systems built for factory layoffs will meet a different kind of job loser. What should change: benefit length, wage insurance or retraining?
What prompted this#
Since October 2025, 86 papers carry an unemployment-insurance, unemployment-benefit, job-search or displaced-worker keyword (66 in economics journals): "unemployment insurance" on 18, "unemployment benefits" on 10, "job search" on 7. Only 20 of them also carry an automation, AI or algorithm keyword (automation 7, artificial intelligence 5, generative AI 4).
Questions#
- Do workers who lose jobs exposed to language models search longer and take larger pay cuts than other displaced workers with similar wages?
- Would wage insurance (topping up a lower-paid new job) get them back to work faster than longer benefits?
- Do job seekers who use AI to write applications get more interviews, and does the flood of applications lengthen everyone's search?
Data that could help#
Administrative unemployment spells with prior occupation (registers in Scandinavia, Germany's IAB data), occupational AI-exposure scores, and job-board application data.
Next step#
Compare unemployment durations and re-employment wages of workers displaced from high and low AI-exposure occupations since 2023 in one country's register data.