V8 methodology
A deterministic, open-data ranking of how exposed occupations are to current AI capabilities, built from the retained V7 research inputs.
What the headline means
An AI Exposure Rank of 72/100 means the occupation ranks higher than approximately 72% of occupations in the Singapore reference market for AI task exposure.
It does not mean a 72% chance of job loss, 72% of jobs disappearing, or 72% of tasks being automated. The score is relative to the other occupations in this release.
How V8 is calculated
1. Build the exposure ensemble
Combine Felten AIOE, Anthropic observed usage, Eloundou GPT exposure and the ILO refined index using published reliability weights. Missing sources are renormalized rather than imputed.
2. Rank occupations within Singapore
Tied values receive their midrank. Index points are 100 × (midrank − 1) / (N − 1), with N = 562. Bands are quintiles: 0–19 Very Low, 20–39 Low, 40–59 Moderate, 60–79 High and 80–100 Very High.
3. Keep employment economics separate
Structural substitution pressure ranks exposure × (1 − human bottleneck). Augmentation
potential ranks exposure × human bottleneck. Official demand, firm adoption, workforce
age, entry-level sensitivity and transition evidence are published beside those indices
rather than hidden inside the headline. Entry-level sensitivity is currently unknown: no suitable open official occupation-level hiring series isolates entry-level workers.
Likely pathway rules
Limited direct change
AI Exposure Rank below 40.
Hiring or substitution pressure
Substitution pressure is High+, directly supported adoption is established or leading, and demand is not strong.
Augmentation-led growth
Augmentation potential is High+ and official demand is strong.
Demand-buffered redesign
Substitution pressure is High+ but official demand is strong.
Workflow redesign
All remaining combinations; evidence supports change but not a narrower employment claim.
Confidence and sensitivity
Confidence is High, Medium or Low based on mapping quality, source coverage, task evidence and policy caps. The site shows the limiting factors; it does not render confidence as a probability.
Sensitivity is the score range produced by equal source weights and leave-one-source-out variants. It is an assumption-sensitivity range, not a statistical confidence interval.
Important limitations
- Exposure measures disagree about constructs and magnitude; the ensemble does not make them causal.
- Detailed Singapore occupation employment is estimated from broader official groups.
- Demand and age evidence can be broader than the individual SSOC occupation.
- Output-demand elasticity, firm reorganization and job quality are not measured as occupation scores.
- Observed employment effects can appear first in hiring, hours, contractors or career ladders rather than layoffs.
- The deduplicated U.S. BLS cross-country check finds essentially no rank association (ρ = 0.01 across 243 unique mapping signatures); it does not validate Singapore employment outcomes.
- Synthetic modern roles are estimates and are kept separate from official occupations.