About This Dashboard

Automation Exposure Analysis

Note: This page was generated with AI and has not been reviewed thoroughly. For deeper technical detail, visit the Dashboard GitHub and review the ARCHITECTURE.md and PRD.md files. You can use AI to help summarize any questions you may have.

Built for Utah's Office of Artificial Intelligence Policy (OAIP) as part of research to explore how AI is impacting the economy. Visualizes AI automation exposure across U.S. occupations, work activities, and time using multiple independent AI scoring datasets.

How Numbers Are Calculated

Task Completion Weight (task_comp)

Every O*NET task has survey-measured frequency, importance, and relevance. The task completion weight determines how much a task contributes to an occupation's total work:

Time method:
task_comp = freq_mean

Value method:
task_comp = freq_mean × relevance × importance

Auto-Aug Multiplier

When enabled, each task's weight is scaled by its AI automatability score (auto_aug_mean, 0–5 scale):

task_comp (with auto-aug) = task_comp × (auto_aug_mean / 5)

% Tasks Affected

The share of total weighted task completion attributable to AI-exposed tasks, relative to the full ECO baseline for that occupation:

% Tasks Affected = Σ(AI task_comp) / Σ(ECO task_comp) × 100

Always a ratio-of-totals, never an average of per-task percentages. The ECO denominator covers all tasks; the AI numerator covers only tasks present in the selected AI dataset(s).

Workers Affected

Workers Affected = (% Tasks Affected / 100) × Employment

Employment figures from BLS OEWS 2024 (national or Utah).

Wages Affected

Wages Affected = (% Tasks Affected / 100) × Employment × Median Annual Wage

Displayed with adaptive units: $B when ≥ $1B, $M when ≥ $1M, $K when ≥ $1K.

Multi-Dataset Combination

When multiple AI datasets are selected, scores are combined per-task before aggregation:

Average: combined_score = mean(score across selected datasets)
Max: combined_score = max(score across selected datasets)

Explorer Metrics

Pre-computed across 8 AI sources (AEI Conv. v1–v4, AEI API v3–v4, MCP Cumul. v4, Microsoft). For each task, up to 8 sources contribute:

Auto Avg↑ / Auto Max↑ (with values):
avg or max of per-task auto_aug scores, only over tasks with ≥ 1 source value

Auto Avg (all) / Auto Max (all):
same, but null scores treated as 0; covers all tasks in the group

Pct Avg↑ / Pct Max↑:
avg or max of per-task pct_normalized (share of AI conversations); only tasks with values

Σ Pct Avg / Σ Pct Max:
sum (not average) of per-task pct_normalized across tasks with values

Group-Level Aggregation

At group levels (Major / Minor / Broad), explorer metrics are computed from the unique task norms pooled across all occupations in the group — not averaged from per-occupation values. Employment and wages are summed across occupations.

Work Activity Employment Allocation

Time (freq) weighting:
emp_allocated = emp × (task freq_mean / Σ freq_mean for occ)

Value (imp) weighting:
emp_allocated = emp × (task freq×rel×imp / Σ freq×rel×imp for occ)

Data Sources

  • O*NET 2025 / 2015 — Task inventory, work activity hierarchy (GWA / IWA / DWA), frequency and importance ratings
  • BLS OEWS 2024 — Employment and median annual wage by occupation, national and Utah
  • Anthropic Economic Index (AEI) — AI automatability scores from Claude conversation analysis across four snapshot dates (Dec 2024 – Nov 2025), plus cumulative versions
  • MCP Server Pipeline — AI task classification via Model Context Protocol server logs, four snapshot dates (Apr 2025 – Feb 2026)
  • Microsoft Occupational AI Analysis — Independent AI exposure ratings across occupations (Sep 2024)