מסמך טכני מיומן המחקר · באנגלית, כפי שנכתב בשלב העבודה
Step 201: AI exposure of Israel's unemployed, by occupation of last job
Sources: CBS annual Labour Force Survey, table 5.3 (unemployed who worked in the past 12 months, by occupation of last job), editions 2017, 2018, 2021 and 2022; employed persons by occupation from ILOSTAT (national source CBS). Exposure: ILO 2025, unweighted mean of two-digit groups per major group; groups 7 and 8 combined as CBS publishes them. Archived in data/processed/israel_unemployed_occupation_v201.json (CI run 41983819513).
Relative risk = share of the unemployed in a group / share of the employed in that group.
| Occupation group (most exposed first) | ILO exposure | RR 2017-18 | RR 2021-22 |
|---|---|---|---|
| Clerical support workers | 0.549 | 1.35 | 1.60 |
| Professionals | 0.400 | 0.62 | 0.60 |
| Managers | 0.362 | 0.82 | 0.74 |
| Technicians and associate professionals | 0.351 | 0.97 | 1.01 |
| Service and sales workers | 0.238 | 1.34 | 1.28 |
| Skilled trades and operators (7-8) | 0.198 | 0.97 | 1.17 |
| Elementary occupations | 0.152 | 1.68 | 1.64 |
Reading:
- The unemployed were slightly less exposed to AI than the employed in both periods (exposure gap -0.016 and -0.014), because professionals are under-represented among the unemployed.
- Clerical support workers, the most exposed group, were already over-represented among the unemployed, and more so in 2021-22 (RR 1.35 to 1.60). The slope of the change in log RR on exposure is +0.17, driven mostly by this group; with seven groups it is not a test.
- All four years precede the public release of generative AI tools (late 2022). A rise in the clerical share of unemployment after 2023 would therefore not by itself show an AI effect: the trend started earlier. The post-AI comparison needs the 2023 and later editions of table 5.3, whose publication folders could not be located automatically.
Limits: the unemployed without a known last occupation are excluded; CBS suppresses small cells (group 6 is dropped); 2021 follows the COVID shock; descriptive only, no causal claim.
Update: the 2023 and 2024 editions were found later; see step 204.