The predictions have been catastrophic. MIT simulations suggested more than one in ten U.S. workers could be readily replaced by AI. Goldman Sachs estimated AI could match humans in 47% of tasks. McKinsey projected 375 million workers would need to change careers.
But here's what's strange: The predicted job losses haven't materialized at the scale forecasted.
The Problem With Automation Forecasts
Historical forecasts consistently relied on theoretical task-replacement models. They ask: "Can AI theoretically do this task?" If yes, they assume displacement.
But theoretical vulnerability rarely translates linearly into actual job losses. Nobel laureate Daron Acemoglu estimates AI will increase GDP by only 1.1% to 1.6% over the next decade.
Introducing Observed Exposure
Anthropic's Economic Index analyzed over two million real-world professional interactions with Claude. The metric is called "observed exposure"—measuring what AI is actually doing in professional environments, not what it theoretically could do.
Key finding: 97% of observed tasks fall into theoretically feasible categories, but the deployment gap is massive.
The 22 Standard Occupational Categories
The analysis maps AI impact across 22 major occupational categories:
- Highest Exposure: Computer & Mathematical (~33% observed vs ~94% theoretical)
- Moderate Exposure: Office/Admin, Legal, Management, Business, Engineering
- Low Exposure: Healthcare Practitioners, Community/Social Service
The Protected 12: Near-Zero AI Exposure
Twelve occupational categories show near-zero observed exposure:
- Healthcare Support: Physical presence, emotional intelligence, hands-on care
- Protective Service: Physical intervention, split-second judgment
- Food Preparation: Physical dexterity, taste, presentation
- Building Maintenance: Physical environment manipulation
- Personal Care: Physical touch, personal relationships
- Farming/Fishing/Forestry: Outdoor physical labor, unpredictable environments
- Construction/Extraction: Physical construction, spatial reasoning
- Installation/Maintenance/Repair: Diagnosis and repair of physical systems
- Production: Physical manufacturing, quality control
- Transportation: Physical goods movement, navigation
The Deployment Gap Explained
For most occupations, the deployment gap exceeds 40 percentage points. This gap represents: quality requirements, legal barriers, integration complexity, trust deficits, and human oversight requirements.
What This Means For Your Career
- Theoretical Vulnerability ≠ Actual Risk
- Physical presence matters
- Human judgment has value
- Tech careers have highest exposure but even there, deployment lags theory significantly
Summary
The AI job displacement narrative has been overstated. Not because AI isn't capable, but because capability doesn't automatically translate to deployment. The future isn't AI versus humans—it's AI integrated with human oversight, judgment, and presence.