AI Ethics & Governance
A cross-national analysis of 2,216 AI policies across 193 countries
About this project
This project analyses 2,216 AI policy documents from 193 countries, drawn from the OECD.AI Policy Observatory, to answer a straightforward question:
Do countries translate their ethical AI commitments into enforceable governance, and how does this vary between high-income and developing countries?
Three complementary lenses — ethics governance depth, UNESCO alignment, and implementation capacity — are used to trace how AI governance performs from the abstract (values and principles) to the concrete (policy actions and institutional arrangements). The central finding is that ethical principles are widely endorsed but rarely operationalised: accountability mechanisms are the most systematically omitted governance feature worldwide.
All work is in progress. Comments and feedback are welcome.
Research volumes
The analysis is presented in four companion volumes:
📘 Book 1 — AI Governance Implementation Capacity
A cross-national analysis of five implementation capacity dimensions (C1–C5) — clarity, resources, enforcement, accountability, and coordination — across 193 countries.
📗 Book 2 — AI Ethics Governance Depth
A systematic analysis of ethics governance depth through five dimensions (E1–E5) — framework depth, rights protection, governance mechanisms, operationalisation, and inclusion.
📕 Book 3 — UNESCO AI Ethics Recommendation Alignment
An assessment of how national AI policies align with the UNESCO Recommendation on the Ethics of AI — 4 values, 10 principles, and 11 policy action areas.
Policy documents (draft)
Two working documents translate the empirical findings into operational guidance for development research. These are draft versions circulated for review and comment.
Stage-by-stage standards, decision prompts, checklists, and templates for research teams using AI tools, structured around the project cycle.
An evidence brief presenting the research findings and analytical framework underpinning the guidance.
Data & code
Analysis code, data processing pipelines, and scoring methodology are available in the repository:
- Repository: github.com/lsempe77/ai-governance-capacity
- Data source: OECD.AI Policy Observatory — 2,216 documents across 193 countries
- Scoring: Three-model LLM ensemble with inter-rater reliability checks
Contact
Lucas Sempé
3ie — International Initiative for Impact Evaluation
lsempe@3ieimpact.org