How We Assess AI Impact on Careers
Transparent methodology for our AI Resilience ratings, including data sources, calculation methods, and honest limitations.
On This Page
Why We Built This
Most AI impact predictions rely on outdated single-source data, like the widely-cited 2013 Oxford study that predicted 47% of jobs were at high risk of automation. While groundbreaking at the time, that research is now over a decade old and didn't anticipate developments like large language models.
We built this system to provide transparent, multi-source assessments that acknowledge uncertainty. Rather than giving a false sense of precision with a single "automation probability" number, we evaluate careers across four dimensions and combine them into categorical ratings.
Our commitment: We update our assessments as new research emerges and are transparent about our methodology's limitations. If you work in an occupation and believe our assessment is wrong, we want to hear from you.
Our 4-Dimensional Framework
We assess AI impact across four dimensions, then combine them into an overall classification.
What it measures
How much of this job's daily work involves tasks that AI can currently perform or assist with.
How we calculate it
We use the AIOE (AI Occupational Exposure) dataset from Felten, Raj, and Seamans (2021), which measures exposure by analyzing how AI capabilities map to occupational abilities defined in O*NET.
- Low: Bottom 33% of exposure scores
- Medium: Middle 34% of exposure scores
- High: Top 33% of exposure scores
Limitations
- Based on task descriptions, not actual workplace AI adoption
- Doesn't distinguish between "AI can do this" and "employers are using AI for this"
- O*NET task descriptions may lag behind how jobs are actually performed
How We Classify Careers
We don't use a simple formula. Instead, we apply a decision tree that weighs different signals based on research about what actually drives job displacement.
AI-Resilient
Meaning: This career has strong protection against AI disruption.
Typical profile: Low task exposure, moderate-to-strong human advantage, stable or growing demand
Examples: Electricians, Plumbers, Nurses, HVAC Technicians
AI-Augmented
Meaning: AI will significantly change how this work is done, but demand for workers remains strong.
Typical profile: High task exposure, but strong growth or strong human advantage
Examples: Software Developers, Graphic Designers, Financial Analysts
In Transition
Meaning: This role is actively evolving — some tasks are being automated while new responsibilities emerge.
Typical profile: High exposure + high automation risk + stable or slow growth
Examples: Customer Service Reps, Paralegals, Medical Coders
High Disruption Risk
Meaning: This occupation faces significant decline due to automation.
Typical profile: High automation risk + declining demand + weak human advantage
Examples: Data Entry Clerks, Telemarketers, Word Processors
Key Decision Rules
- 1. Job growth is our strongest signal. Rapidly declining occupations with high automation risk are classified as High Disruption Risk.
- 2. Strong human advantage provides protection. Even high-exposure jobs can be AI-Resilient if they require irreplaceable human skills.
- 3. We err toward caution. When signals conflict, we assign the more conservative rating.
What We Don't Know
We believe in epistemic humility. Here's what our methodology cannot capture:
Inherent Uncertainty
- AI capabilities are changing rapidly. A job rated "AI-Resilient" today could face disruption from a breakthrough we can't predict.
- Predictions ≠ reality. All models are wrong; some are useful. Our ratings are educated estimates, not guarantees.
Data Gaps
- Small occupations have noisy data. BLS projections for occupations with fewer than 50,000 workers are less reliable.
- New occupations aren't captured. Jobs created in the last few years lack historical data.
- Geographic variation. AI adoption differs by region; our national-level scores may not match your local market.
What We Can't Measure
- Your specific employer. A tech-forward company and a traditional firm may automate very differently.
- Policy changes. Regulations could accelerate or slow AI adoption.
- Economic shocks. Recessions, pandemics, and other disruptions aren't in our model.
Data Sources
| Source | What we use it for | Last updated |
|---|---|---|
| BLS Employment Projections | Job growth outlook | 2024-2034 projections |
| BLS Occupational Employment Statistics | Wage data | May 2024 |
| O*NET 30.1 | Task descriptions, work activities | 2024 |
| AIOE Dataset (Felten et al.) | AI occupational exposure scores | 2021 |
| EPOCH Framework (Loaiza & Rigobon) | Human advantage scoring | 2024 |
Questions or Corrections?
If you have questions about our methodology or believe an occupation is misclassified, we want to hear from you.