Why this map exists
Global AI governance conversations are increasingly moving from "how powerful is AI?" to "who benefits from AI?", and yet most datasets and visualisations that answer this question originate in, and centre, the Global North.
The A4EAI AI Equity World Map is an attempt to make the structural nature of the global AI divide visible, not as a polemic, but as a policy tool. It is designed to serve researchers, policymakers, civil society organisations, and advocates who need a starting point for understanding where intervention is most needed.
Version 1 of this map uses a mix of primary data (Oxford Insights GARI 2024, which is peer-reviewed) and A4EAI composite scores that have not yet been peer-reviewed. The composite scores represent our best current synthesis of available data, and all weighting decisions are documented on this page. We encourage scrutiny, replication, and critique.
Four dimensions of AI equity
A key design principle is that AI equity is not one thing. A country may have strong policy governance but no compute infrastructure. Another may have high internet access but produce no AI models. We deliberately separate four distinct layers to avoid conflating them.
This is the only dimension based entirely on a primary, peer-reviewed dataset. Scores are taken directly from the Oxford Insights Government AI Readiness Index 2024, which examines 40 indicators across three pillars: Government, Technology Sector, and Data & Infrastructure. The Oxford GARI is cited by UNESCO and the G20 as an authoritative benchmark.
| Sub-pillar | Weight | What it measures |
|---|---|---|
| Government pillar | 33% | AI strategy, governance & ethics, digital capacity |
| Technology sector | 33% | Innovation capacity, human capital, market maturity |
| Data & infrastructure | 33% | Data availability, representativeness, infrastructure |
AI Access is an A4EAI composite score. It distinguishes between connectivity (internet access) and AI tool reachability, recognising that a country may have reasonable internet penetration but still face barriers: cost, language, political restriction, or device limitations. This is a distinction most indices collapse.
| Indicator | Weight | Source |
|---|---|---|
| Internet penetration | 30% | ITU 2024 |
| Mobile broadband coverage | 20% | GSMA Intelligence |
| Data affordability (% income) | 20% | Alliance for Affordable Internet |
| AI tool political access | 15% | A4EAI assessment (Freedom House + tool availability) |
| Local language AI availability v2 | 15% | Masakhane, GhanaNLP, AI4Bharat datasets |
The AI tool political access sub-score (15%) is currently a qualitative A4EAI assessment based on Freedom House internet freedom data combined with known tool restrictions. It will be replaced by a quantitative index in v2.
This dimension deliberately captures both government-led governance (national AI strategies, legislation) and civil society participation, recognising that a government can have an AI strategy with zero meaningful public input. Both are important and scored separately.
Crucially, this dimension also scores authoritarian AI deployment and surveillance AI as negative governance factors, a country that deploys AI for citizen surveillance scores lower on governance regardless of how sophisticated its strategy documents are.
| Indicator | Weight | Source |
|---|---|---|
| National AI strategy quality | 25% | OECD AI Policy Observatory |
| AI ethics framework adoption | 20% | UNESCO RAM assessments |
| Civil society participation score | 20% | A4EAI assessment |
| Regulatory enforcement capacity | 20% | Oxford GARI governance pillar |
| Surveillance AI penalty new | −15% | Freedom House + AI Now Institute |
This is A4EAI's most distinctive dimension and the one most directly tied to our advocacy thesis: the Global South is being incorporated into AI as a market and a data source, but excluded from authorship. The Co-creator Index measures the degree to which a country is a producer of AI, not just a user of tools built elsewhere.
| Indicator | Weight | Source |
|---|---|---|
| AI research paper output | 25% | CSET / Stanford HAI |
| Foundation models produced | 25% | Epoch AI model database |
| AI standards body participation | 20% | ISO/IEC JTC1, ITU-T, GPAI membership data |
| Domestic AI company ecosystem | 15% | Stanford HAI private investment data |
| Local language model existence new | 15% | Masakhane, GhanaNLP, AI4Bharat, HuggingFace |
How scores are calculated
All dimensions are normalised to a 0–100 scale. Higher scores indicate greater equity, meaning better access, stronger governance, more co-creation capacity. The map does not produce a single composite "AI equity score" for each country, precisely because collapsing all four dimensions into one number would obscure the structural differences between them.
score = Σ (indicator_value × indicator_weight)
# Where each indicator_value is normalised 0–100:
normalised = (raw_value − min_value) / (max_value − min_value) × 100
# For the AI Readiness dimension:
score = Oxford_GARI_score # direct, no transformation
# Missing data handling:
missing → regional_peer_group_mean imputation
# (same method as Oxford GARI 2024, documented pp.38–39)
Where country-level data is unavailable for a sub-indicator, we use the regional peer-group mean, the same approach as Oxford GARI 2024 (see their methodology, pp.38–39). Countries where more than 40% of sub-indicator data is missing are shown as "No data" rather than imputed.
A score of 80+ means strong capability, access, or governance relative to the global distribution. A score of 20 or below indicates severe structural barriers. Scores are relative, not absolute, they reflect position within the current global distribution, which itself reflects historic inequities.
Primary data sources
Every score on this map can be traced to one or more of the following sources. We distinguish between primary sources (peer-reviewed or institutionally validated datasets) and secondary sources used for triangulation.
| Source | Used for | Type | Update frequency |
|---|---|---|---|
| Oxford Insights GARI 2024 primary | AI Readiness (100%), governance pillar input | Peer-reviewed index | Annual (Dec) |
| ITU ICT Development Index | AI Access, connectivity layer | UN agency dataset | Annual |
| Stanford HAI AI Index 2024 | Co-creator, research output, investment | Academic index | Annual (Apr) |
| CSET AI research database | Co-creator, paper output by country | Academic dataset | Quarterly |
| OECD AI Policy Observatory | AI Governance, strategy quality | Intergovernmental | Continuous |
| UNESCO Readiness Assessment Methodology | AI Governance, ethics framework adoption | UN agency dataset | Annual |
| GSMA Mobile Intelligence | AI Access, mobile coverage | Industry dataset | Quarterly |
| Alliance for Affordable Internet | AI Access, affordability | Civil society index | Annual |
| Epoch AI model database | Co-creator, models produced by country | Research dataset | Continuous |
| Freedom House Internet Freedom | AI Access (political), Governance (surveillance) | Civil society index | Annual |
| HuggingFace model registry v2 | Local language model availability | Open dataset | Continuous |
| Masakhane / GhanaNLP / AI4Bharat v2 | African & South Asian language AI coverage | Research communities | Ongoing |
Known limitations
We believe methodological honesty strengthens rather than undermines advocacy. The following limitations are known and actively being addressed in the Version 2 roadmap.
Composite scores not yet peer-reviewed
Three of four dimensions (Access, Governance, Co-creator) are A4EAI composites. Only the AI Readiness dimension uses a fully peer-reviewed primary source (Oxford GARI). A methodology paper is in preparation.
Missing data disproportionately affects the Global South
Countries with the fewest data points are often those most affected by AI inequity. Regional imputation reduces this bias but does not eliminate it.
Civil society participation is qualitative
Our civil society sub-score currently relies on A4EAI assessments without a fully quantified rubric. A structured qualitative framework is under development with regional partners.
Local language AI not yet quantified
The local language model availability indicator is flagged as "v2" because no standardised cross-country dataset currently exists. We are in dialogue with Masakhane and GhanaNLP to address this.
Subnational variation is invisible
Country-level aggregation hides enormous internal inequity, for example, Nigeria's Lagos tech hub versus its rural north. Future versions will explore subnational layers where data exists.
Gender equity not yet a standalone dimension
Gender equity in the AI workforce is not currently a scored dimension. This is a significant gap. We are sourcing data from ILO, UNESCO, and EQUALS to add this in v2.
What's next
Sought data & advisory partners
To move from Version 1 to credible policy infrastructure, A4EAI is actively seeking partnerships with the following types of organisations. Outreach is underway.
If your organisation works on AI equity, local language AI, digital access, or AI governance in the Global South and would like to contribute data, validate scores, or join the advisory board, please contact research@a4eai.com.
Version history
How to cite this work
This map and its composite scores are published under a Creative Commons Attribution-ShareAlike 4.0 licence. You are free to use, adapt, and share with attribution.
Retrieved from https://www.a4eai.com/ai-equity-map
# For the AI Readiness dimension specifically, also cite:
Oxford Insights. (2024). Government AI Readiness Index 2024.
https://oxfordinsights.com/ai-readiness (CC BY-SA 4.0)