Global AI Dataset (GAID) Project: GAID AI Development Index

Construction

All values come from the latest version of the GAID dataset (harmonising global AI data from 11 verified international sources, covering 227 countries and territories); no modelled or imputed values anywhere. Six formative pillars, namely (1) Research & Innovation; (2) Talent & Skills; (3) Governance & Regulation; (4) AI Economy & Investment; (5) Infrastructure & Compute; (6) Responsible AI & Society, are designed and built to calculate the GAID AI Development Index (applying equal weighting).

Scores are computed per annual edition, where each component contributes its latest observation within a three-year lookback window, and its vintage is recorded. Heavy-tailed counts are log-transformed, winsorised at the 1st/99th percentiles, then min–max scaled to 0–100 within each edition. Here, scores measure relative standing among the pool of included countries. A pillar is scored only when at least half its components are present/available. Moreover, the overall GAID AI Development Index requires at least four of six pillars.

Robustness Check—GAID w1 v2 dataset (Edition 2025)

Weighting. Equal-weight and PCA-derived scores correlate at ρ = 0.95–1.00 across pillars, suggesting that the transparent equal-weight choice is empirically indistinguishable from the data-driven alternative.

Normalisation sensitivity. Rankings correlate at ρ = 0.92–1.00 across min–max, z-score and percentile variants.

External validity. Against the fully held-out Tortoise Global AI Index: overall ρ = 0.82; Research 0.76, Talent 0.87, Commercial 0.73. Divergences on the government-strategy and infrastructure pillars represent different constructs.

Internal consistency. Reflective component groups reach Cronbach’s α = 0.96 (GovTech; GIRAI). Pillars are formative composites, so cross-facet α is informational.

Data Availability & Reuse

The GAID dataset is published on Harvard Dataverse and updated annually; this dashboard rebuilds automatically from the latest wave. The pipeline (sync → harmonise → screen → indices → site) is open-source Python with per-wave validation reports. It is noteworthy that responsible-AI components incorporate dimension scores from the Global Index on Responsible AI with attribution.

Remark

The full paper disclosing and detailing the open-source methodology will be published in due course.