1. The 15 AI Capability Dimensions
To avoid evaluating “AI” as a monolithic entity, JobsVsAI maps occupational workflows against fifteen discrete capability dimensions spanning four structural domains:
Cognitive & Analytical
- Complex Reasoning & Deduction
- Quantitative & Mathematical Computation
- Document & Data Synthesis
- Code & Algorithmic Generation
- Creative Pattern Synthesis
Perceptual & Spatial
- Visual Scene Recognition & Interpretation
- Spatial Awareness & Environmental Navigation
- Audio & Speech Processing
- Multi-Modal Sensor Integration
Physical & Manipulation
- Fine Motor Dexterity & Tool Operation
- Gross Motor Coordination & Physical Stamina
- Real-Time Physical Manipulation in Unstructured Settings
Social, Emotional & Governance
- Empathetic Interpersonal Communication
- High-Stakes Ethical & Strategic Judgement
- Accountability, Liability & Regulatory Governance
2. Task-Level Geometric Mean & Bottleneck Caps
Standard linear averaging fails in occupational risk modeling because strong capability in text synthesis cannot compensate for zero dexterity in physical surgery.
For every task \(T\), capability match across each dimension \(i\) is computed using a logistic capability margin curve against commercial AI frontier capabilities. Overall task capability fit is aggregated via a weighted geometric mean with critical bottleneck caps:
Critical Bottleneck Caps: When a task requires high capability in a dimension where commercial AI exhibits a severe shortfall, a deterministic bottleneck cap is enforced, bounding the task score regardless of strength in other dimensions.
3. Task Weighting and AI Exposure Aggregation
Individual task scores are aggregated into occupational headline scores using O*NET 30.3 empirical weights. Each task statement is scaled by its surveyed importance and frequency across the verified occupation:
4. Structural Friction & Replacement Risk Modeling
AI Exposure and Replacement Risk diverge because technical capability is subject to real-world friction before human labour can be substituted. JobsVsAI combines task-level automation feasibility with four structural friction layers:
- Environmental & Physical Dependency: Requirements for physical presence, dexterity, tool handling, and unconstrained spatial navigation in dynamic environments.
- Human Accountability & Trust: Legal liability, fiduciary responsibility, ethical sign-off, patient/client rapport, and high-consequence decision-making requiring an accountable human party.
- Adoption Economics & Integration Friction: Capital costs of enterprise deployment, system integration complexity, regulatory compliance timelines, and organizational workflow inertia.
- Labour-Market Resilience: Macroeconomic workforce elasticity, demographic shortages, institutional certification barriers, and wage dynamics.
5. Preliminary Estimation Framework (E1, E2, E3)
For occupations that have not completed full individual task-level analysis, JobsVsAI produces provisional estimates bounded by empirical confidence tiers:
Point Estimate
High direct task coverage (>80%) from mapped shared work activities.
Range Estimate
Partial task coverage (30%–80%) with bounded upper and lower uncertainty intervals.
Cluster Proxy
Structural proxy mapping derived from nearest-neighbor verified occupational clusters.
