Authors

Peter Martey Addo

Thomas Melonio
Anastesia Taieb
Laura Landrein
Coordination
Research
Thomas Melonio (AFD)
Peter Martey Addo (AFD)
papers

Development
AI Investment

Opportunities
for Sustainable
Potential Index:

No. 330
Mapping Global

NOVEMBER 2024
Agence Française de Développement

Papiers de recherche Research Papers

Les Papiers de Recherche de l’AFD ont pour but de AFD Research Papers are intended to rapidly
diffuser rapidement les résultats de travaux en cours. disseminate findings of ongoing work and mainly
Ils s’adressent principalement aux chercheurs, aux target researchers, students and the wider academic
étudiants et au monde académique. Ils couvrent community. They cover the full range of AFD work,
l’ensemble des sujets de travail de l’AFD : analyse including: economic analysis, economic theory, policy
économique, théorie économique, analyse des analysis, engineering sciences, sociology, geography
politiques publiques, sciences de l’ingénieur, and anthropology. AFD Research Papers and other
sociologie, géographie et anthropologie. Une publications are not mutually exclusive.
publication dans les Papiers de Recherche de l’AFD
n’en exclut aucune autre. The opinions expressed in this paper are those of the
author(s) and do not necessarily reflect the position of
Les opinions exprimées dans ce papier sont celles de AFD. It is therefore published under the sole
son (ses) auteur(s) et ne reflètent pas nécessairement responsibility of its author(s).
celles de l’AFD. Ce document est publié sous l’entière
responsabilité de son (ses) auteur(s).
AI Investment Potential Index Abstract Keywords
This paper examines the AI Investment Potential Index,
Mapping Global Opportunities potential of artificial intelligence sustainable development,
for Sustainable Development (AI) investment to drive artificial intelligence,
sustainable development across investment decisions,
diverse national contexts. By equitable growth,
evaluating critical factors, AI readiness.
including AI readiness, social
inclusion, human capital, and JEL codes
macroeconomic conditions, we O33 Technological Change
construct a nuanced and Choices and Consequences
AUTHORS comprehensive analysis of the Diffusion Processes
global AI landscape. Employing
F63 Economic Development
advanced statistical techniques
Peter Martey ADDO
and machine learning C43 Index Numbers
AI Lead
algorithms, we identify nations and Aggregation
Head of Emerging Tech Lab
with significant untapped G11 Portfolio Choice
Agence Française de
potential for AI investment. Investment Decisions
Développement (AFD)
France Q01 Sustainable Development
We introduce the AI Investment
Potential Index (AIIPI), a novel Acknowledgements
Thomas MELONIO 1
instrument designed to guide We would like to express our
Chief Economist
financial institutions, heartfelt gratitude to Alix Le
Executive Director of Innovation
development banks, and Coguic for her exceptional
Strategy & Research at Agence
governments in making contributions to the data
Française de Développement
informed, strategic AI collection efforts for this study.
(AFD)
investment decisions. The AIIPI Her dedication, attention to
France
synthesizes metrics of AI detail, and meticulous work were
readiness with socio-economic indispensable to the success of
Anastesia TAIEB indicators to identify and this research.
Research Assistant & Innovation highlight opportunities for We also extend our sincere
Officer (AFD) fostering inclusive and appreciation to Antoine Godin
sustainable growth. (AFD), Laurent Giuseppe
Laura LANDREIN D’Aronco (UNECA), and the
Research Assistant & Innovation The methodological novelty lies Organizing Committee of the
Officer (AFD) in the weight selection process, 12th IMF Statistical Forum. Their
which combines statistical valuable feedback, thoughtful
modeling and also an entropy- insights, and constructive
based weighting approach. critiques significantly enhanced
Furthermore, we provide the quality of this manuscript.
detailed policy implications to Their expertise and careful
support stakeholders in making review played a crucial role in
COORDINATION
targeted investments aimed at refining and elevating the work
reducing disparities and to its current form.
Thomas MELONIO (AFD)
advancing equitable
Peter Martey ADDO (AFD) technological development. Accepted
November 2024

1
Corresponding Author:
Email address: meloniot@[Link]
Résumé L'AIIPI combine des indicateurs Mots-clés
Cet article examine le potentiel de préparation des pays à l'IA et Indice du Potentiel
d’ investissement en intelligence des données socio­ d'Investissement en IA,
artificielle (IA) pour stimuler le économiques pour identifier des développement durable,
développement durable dans opportunités de croissance intelligence artificielle,
divers contextes nationaux. En inclusive et durable, en décisions d’investissement,
évaluant des facteurs clés tels analysant des facteurs clés tels croissance équitable,
que l’état de préparation à l’IA, que les infrastructures état de préparation à l’IA.
l’inclusion sociale, le capital numériques, le développement
humain et les conditions technologique, l'ouverture à l'IA,
macroéconomiques, nous la stabilité politique et
élaborons une analyse nuancée l'attractivité économique.
et complète du paysage L’innovation méthodologique de
mondial de l’IA. cet indice réside dans le
processus de sélection des
À l’aide de techniques pondérations, qui combine une
statistiques avancées et modélisation statistique et une
d’algorithmes d’apprentissage approche de pondération basée
automatique, nous identifions les sur l’entropie. Enfin, nous
pays présentant un potentiel proposons des
inexploité significatif pour les recommandations détaillées en
investissements en IA. Nous matière de politique publique
introduisons l’Indice de Potentiel pour accompagner les parties
d’Investissement en Intelligence prenantes dans la réalisation
Artificielle (AIIPI), un outil d'investissements ciblés, visant à
novateur conçu pour orienter les réduire les disparités et à
institutions financières, les promouvoir un développement
banques de développement et technologique inclusif et
les gouvernements dans leurs durable.
décisions d’investissement
stratégique en IA.
1. Introduction

As we reach the midpoint of the 2030 for explicit or implicit objectives, infers from
Agenda for Sustainable Development, its inputs how to generate outputs such as
initially established to achieve the predictions, content, recommendations, or
Sustainable Development Goals (SDGs), the decisions that can influence physical or
global community confronts a daunting virtual environments" (OECD, 2024). AI is
reality: progress has significantly lagged already driving substantial changes across
behind expectations. Currently, only 17% of industries. For example, AI is optimizing
the SDGs are on track, with nearly half (48%) energy management, enhancing the
facing considerable delays, and more than efficiency of renewable resources,
a third (35%) either stagnating or regressing personalizing education to address
compared to their 2015 baselines (United learning gaps, and revolutionizing
Nations Statistics Division, 2024). This healthcare diagnostics and prognostics
troubling scenario results from a complex (Addo et al., 2021; Nahar, 2024; Willige, 2024).
interplay of factors, including climate
change, geopolitical tensions, and global From an economic perspective, the impact
health crises, which have collectively of AI remains uncertain, with projected
undermined international cooperation and contributions from generative AI ranging
impeded progress (United Nations from as low as 0.5% of global GDP over the
Environment Programme, 2023; Le et al., next decade (Acemoglu, 2024) to as high as
2022; Kharas, 2021). 7%, or $7 trillion (Goldman Sachs, 2023). Such
varying estimates reflect the current
Amid these challenges, the rapid evolution ambiguity surrounding AI's aggregate
of artificial intelligence (AI) presents both a economic impact, whether through
remarkable opportunity and a formidable recognized gains in existing industries or
challenge. Generative AI technologies such anticipated future benefits across broader
as ChatGPT, Copilot, Gemini, and Claude economic activities.
exemplify AI's potential to transform
economic, social, and environmental However, the potential benefits of AI are
systems. Defined by the Organisation for unevenly distributed. To fully harness AI for
Economic Co-operation and Development sustainable development, substantial
(OECD) as "a machine-based system that, investments are necessary, particularly in

4
developing countries, where access to Similarly, initiatives like the Agence
technology, financing, and skilled labor Française de Développement (AFD) AI
remains limited. This disparity raises critical challenge for marine biodiversity reflect the
questions about the capacity of these growing interest in AI's role in achieving the
nations to engage with and benefit from the SDGs.
AI revolution. For instance, in Africa, funding
remains a significant barrier for AI start-ups In this context, understanding how and
and research initiatives such as where to invest in AI becomes essential for
Masakhane, which focuses on natural fostering inclusive and sustainable growth.
language processing (NLP) research in This paper introduces the AI Investment
African languages (Madu et al., 2024). These Potential Index (AIIPI), a novel tool designed
initiatives require substantial investments to guide development financial institutions,
in cloud computing, skilled personnel, and banks, and governments in making
computational infrastructure to succeed. informed decisions about AI investments.
By integrating AI readiness indicators, social
Recognizing AI's transformative potential, inclusion metrics, and macroeconomic
various development organizations are outlooks, the AIIPI framework provides a
increasingly focusing on strategic multi-faceted assessment of the global AI
investments to address pressing social and landscape, effectively identifying countries
environmental challenges. The AI for with substantial untapped investment
Development (AI4D) program, launched at potential. Through a combination of
the UK AI Safety Summit in November 2023, statistical techniques and advanced
exemplifies this commitment through a machine learning algorithms, this research
collaborative initiative involving the uncovers latent patterns of similarity
International Development Research among countries, thereby highlighting
Centre (IDRC), the Foreign, Commonwealth strategic opportunities for investments
and Development Office (FCDO), the Bill and aimed at bridging current technological
Melinda Gates Foundation (BMGF), and the gaps and fostering equitable growth in AI
United States Agency for International capabilities.
Development (USAID). With CAD 130 million
committed over five years, AI4D aims to This paper is structured as follows: Section 2
leverage AI to reduce inequalities and provides a review of the relevant literature,
strengthen health, education, and food Section 3 outlines the data and methodo­
systems while enhancing resilience to logy used, Section 4 presents the analysis
climate change (International and key findings, and Section 5 discusses
Development Research Centre, 2023). the insights and policy implications.

5
2. Literature Review

Artificial Intelligence (AI) has emerged as a transformative force with the potential to
fundamentally reshape global economies and drive sustainable development. The rapid
advancements in AI technologies, coupled with increasing investments, underscore its
growing significance in driving economic growth and social progress. This literature review
examines the key factors influencing AI adoption and its implications for sustainable
development, emphasizing the diverse challenges and opportunities across different
countries.

The Economic Impact of AI

AI's potential to reshape the global economy has frequently been likened to a new industrial
revolution. McKinsey estimates that generative AI alone could contribute between
$2.6 trillion and $4.4 trillion to global productivity—an amount roughly equivalent to the entire
GDP of the United Kingdom (Chui et al., 2023). However, AI's impact is not uniform across
different economies. The current focus on automating repetitive tasks suggests that while
AI's contribution to productivity could be significant, its broader macroeconomic effects
may remain moderate, resulting in an increase in Total Factor Productivity (TFP) of only 0.55%
to 0.71% (Acemoglu, 2024). TFP, a measure of economic efficiency, reflects how effectively
labor and capital are utilized, incorporating factors such as technology, innovation, and
workforce skills.

Digitalization and AI’s Role in Sustainable Development

The digital economy is increasingly integral to global economic dynamics, contributing 15.5%
of global GDP in 2023 (Al Yahya, 2023). AI is expected to be a key driver of this growth, with
more than two-thirds of new value projected to arise from digitally enabled platforms over
the next decade (World Economic Forum, 2024). The private sector's enthusiasm for AI
investment largely stems from its potential to enhance productivity and drive economic
expansion.

AI's contribution to the United Nations' Sustainable Development Goals (SDGs) is also
significant, with potential impacts—both positive and negative—on 134 out of 169 SDG targets
(Vinuesa, 2020). AI can advance goals related to Industry, Innovation, and Infrastructure
(SDG 9), Reduced Inequalities (SDG 10), Peace, Justice, and Strong Institutions (SDG 16), and
Partnerships for the Goals (SDG 17). However, these opportunities come with substantial risks.

6
For example, while AI-driven data analysis can enhance governance transparency and
accountability, it may also exacerbate inequalities if implemented without sufficient
attention to inclusivity (Nahar, 2024).

The United Nations has already launched over 408 AI-related projects across all 17 SDGs,
highlighting the global commitment to leveraging AI for sustainable development
(International Telecommunication Union, 2023). Nonetheless, the effective realization of AI's
potential requires a nuanced understanding of its applications and the socio-economic
contexts in which it is deployed.

Key Factors Influencing AI Adoption

AI adoption is a priori influenced by several critical factors, including digital infrastructure,

technological sector development, government openness to AI, and political stability. These
factors vary widely across countries, shaping the readiness and attractiveness of AI
investments. In this research, we will test the statistical significance of these families of
variables, and build our index based on these estimates. We will use the number of AI
investments as a proxy of AI adoption in any particular country.

Digital Infrastructure: Robust digital infrastructure may be necessary to unlock AI's

potential. This includes reliable hardware, cloud computing capabilities, high-speed
broadband, and access to quality data (Oxford Insights, 2023). Many developing regions
face significant infrastructural deficits, contributing to a persistent global digital divide. With
over 2.6 billion people lacking internet access, particularly in regions such as Sub-Saharan
Africa, disparities within and between countries continue to impede equitable AI adoption
(United Nations, 2024). Addressing these gaps through targeted investments in digital
infrastructure is therefore essential for inclusive AI growth.

Technological Sector Development: A well-developed technological sector, supported by a

skilled workforce, could also be crucial for AI advancement. High-income countries,
particularly the United States and China, lead in AI development due to substantial
investments, vibrant AI ecosystems, and skilled labor. In 2020, US venture capital accounted
for 43% of global AI investments, followed by China with 20% (OECD, n.d.). These nations have
established themselves as leaders in AI research and commercialization, while many

7
developing countries continue to face significant shortages of AI expertise (Khanal et al.,
2024). There are, however, exceptions, such as Malaysia and the some of the BRICS 2 countries
(Brazil, Russia, India and China), which have made notable progress in AI research and
education (Oxford Insights, 2023). Building a skilled workforce and fostering a supportive
policy environment are essential for countries aspiring to develop a responsible AI
ecosystem.

Government Openness to AI: Government support and regulation may also influence AI
adoption. Effective regulatory frameworks can help mitigate risks associated with AI, such
as bias, discrimination, and privacy violations (Fraisl et al., 2024). The AI Government
Readiness Index by Oxford Insights (2023) ranks countries based on their AI strategies,
regulatory environments, and digital capacities. Strategic policies have enabled Singapore,
for example, to position itself as a global leader in AI readiness despite its small geographic
size (Khanal et al., 2024). Development practitioners could therefore support governments in
formulating AI strategies, implementing ethical frameworks, and strengthening digital
capacities to foster responsible AI ecosystems.

Political Stability and AI Investments: Political stability and economic attractiveness are
likely determinants of AI investment decisions. A stable political climate attracts and retains
investments, talent, and businesses, facilitating AI development. Key indicators such as
Foreign Direct Investment (FDI), government effectiveness, and corruption control are
critical to a country’s investment climate (Kaufmann & Kraay, 2023). Conversely, political
instability can deter investments and foster irresponsible AI applications, leading to adverse
effects such as misinformation and harmful content proliferation.

The existing literature underscores the importance of aligning AI investments with a

country's stage of development. While advanced economies should focus on AI innovation
and regulatory frameworks to maximize benefits, emerging and developing economies
must prioritize foundational infrastructure and skill development. This study introduces the
AI Investment Potential Index (AIIPI) as a novel tool to guide investment decisions,
emphasizing the need for targeted investments to bridge existing gaps and promote
inclusive, sustainable growth. By offering a comprehensive understanding of the global AI
landscape, this research provides a roadmap for leveraging AI's potential to drive global
progress.

2
BRICS is an intergovernmental organization that aims to increase the economic and political influence of its
member countries. BRICS initial membership consists of Brazil, Russia, India, China, and South Africa. In 2024,
Saudi Arabia, Iran, Ethiopia, the United Arab Emirates, and Egypt joined BRICS.

8
3. Data and Methodology

The AI Investment Potential Index (AIIPI) framework is constructed using a comprehensive

range of indicators derived from publicly available datasets such as those provided by the
World Bank, Oxford Insights, GSMA, and Commission nationale de l’informatique et des
libertés (CNIL). These datasets encapsulate multiple facets of AI investment potential,
encompassing governance, infrastructure, human capital, economic environment, and
digital capacity. A detailed enumeration of the variables, including descriptions and data
sources, is available in Annex 1.

Exploratory data analysis, including descriptive statistics and assessment of missing values,
was conducted to refine the dataset. Variables with extensive missing data were excluded,
while highly correlated variables were consolidated to reduce redundancy (see Figure 1).

The principal indicators utilized to develop the AI Investment Potential Index (AIIPI) are as
follows: Access to Electricity, the GSMA Connectivity Index, and the Telecommunication
Infrastructure Index, which represent the foundational infrastructure necessary for the
deployment of AI technologies. Economic measures such as Log GDP per Capita PPP and
Population assess a country’s market potential and financial capacity to support AI growth.
Governance-related indicators, including Government Effectiveness, Political Stability,
and Voice and Accountability, evaluate the quality of public services, political climate, and
regulatory framework, all of which are crucial for long-term AI investments. The Human
Capital Index measures the availability of skilled talent, while innovation capacity is
reflected in the Number of Research Articles and Statistical Data Capacity, representing a
country’s research ecosystem and data management capabilities. Additionally, the
presence of a national AI Strategy and the Data Privacy and Protection Score highlight a
government’s commitment to fostering AI development and ensuring data security.

These indicators, collectively, provide a comprehensive framework for evaluating a

country's AI investment potential by addressing critical factors related to infrastructure,
governance, economic capacity, human capital, and innovation.

9
3.1. Scaling and Normalization

All indicators were normalized using min-max scaling, transforming each metric to a range
of [0, 100] to allow comparability across indicators. In instances of high correlation between
indicators, such as GDP per Capita and GDP per Capita PPP, only the more informative
measure (GDP per Capita PPP) was retained.

Figure 1. Correlation Plot on Scaled data

10
The formula used for scaling is:

(X − Xmin )
Scaled Value = ∗ 100
(Xmax − Xmin )

Where X represents the original value, Xmin is the minimum observed value, and Xmax is the
maximum observed value for the variable.

Variables with missing values (NA) were rescaled by excluding these values when
determining the minimum and maximum. If all values were missing, no scaling was applied,
and the original values were retained. When all non-missing values were identical (i.e., Xmin =
Xmax ), all observations were assigned a value of 100 to avoid division by zero. Table 3 (See
Annex 2) shows the minimum and maximum value of the unscaled variables, Table 1 displays
the descriptive statistics of the scaled variables and Figure 1 represents the correlation plot
on scaled data.

11
 Table 1. Descriptive Statistics of Scaled Variables

Variable Mean Median SD IQR Skewness Kurtosis ShapiroP3 Missing

[Link] 1.75 0.095 9.29 0.617 9.74 101 0 0

[Link] 60.6 100 46.4 100 -0.433 1.31 0 6

[Link] 93.3 100 19.4 0.331 -3.23 12.5 0 8

[Link] 62.6 64.4 24.3 37.8 -0.51 2.49 0 10

[Link]
ore 72.4 100 32.5 50 -0.759 2.52 0 0

[Link] 60.1 62.7 20.3 27.6 -0.51 2.8 0 4

[Link] 52.3 49.2 20.9 31.8 0.14 2.29 0.009 8

[Link] 71.2 74.1 19.1 24.8 -1.09 4.08 0 8

Telecommunication.

[Link] 62 64.9 24.1 34.7 -0.548 2.46 0 8

[Link] 63.2 65.5 20.6 30.4 -0.502 2.82 0 8

Population 4.25 0.791 13.3 2.87 6.42 45.7 0 2

[Link] 54.3 56.4 27 44.9 -0.274 1.97 0 8

[Link] 3.66 0.624 11.5 2.28 6.5 49.8 0 4

[Link] 69.9 74.3 20.9 35.2 -0.612 2.59 0 9

3
Note: If the Shapiro-Wilk p-value (ShapiroP) is greater than 0.05, it suggests that the variable is likely normally
distributed. Shapiro-Wilk Test is more effective for smaller datasets (typically < 5000 observations).

12
3.2. Model Development and Weighting Strategy

The "incoming AI investment counts" indicator was selected as the target variable to assess
the attractiveness of countries for AI investments. This variable was linked to a carefully
selected subset of 13 key indicators, including AI strategy, political stability, GSMA
connectivity, and statistical data capacity. Data from 2020 and 2022 were analyzed to
identify key factors that most strongly predict AI investment potential. This analysis provided
a solid foundation for model evaluation and informed the development of the AI Investment
Potential Index (AIIPI), with the 2022 data playing a central role in constructing the final index.

To construct the AIIPI, we considered three modeling approaches along with an entropy-
based weighting method (Roszkowska et al., 2024) to determine the optimal weightings for
each indicator. Linear models were initially applied to establish a baseline understanding of
the relationships between the indicators and the target variable, highlighting which
predictors had significant associations. These models provided essential insights into linear
interactions between variables. Building on this, Elastic Net Regression was employed to
address multicollinearity among the indicators. By combining L1 (lasso) and L2 (ridge)
regularization techniques, Elastic Net balanced feature selection with model stability. A grid
search was utilized to optimize key hyperparameters—penalty (lambda) and mixture
(alpha)—while five-fold cross-validation ensured robust model performance and minimized
overfitting.

Additionally, Random Forests, a non-linear ensemble learning technique, was implemented

to capture complex interactions between the indicators. This method was particularly useful
for identifying the most important features across a wide range of socio-economic
variables. The Random Forest model was tuned to find the optimal number of predictors at
each split and the minimum number of observations in terminal nodes. Cross-validation
was employed, and the model selection was based on minimizing Root Mean Square Error
(RMSE) to ensure accuracy in predicting AI investment potential.

The model development process followed a structured pipeline. The dataset was split into
training (80%) and testing (20%) subsets to ensure generalizability. During data
preprocessing, missing values were imputed using median replacement for numeric
predictors, preserving dataset integrity and avoiding potential biases. Hyperparameter
tuning for both the Elastic Net and Random Forest models was conducted via grid search
techniques—grid_max_entropy for Elastic Net and grid_regular for Random Forest—to
systematically explore the parameter space and assess the impact of different

13
configurations on model performance. Five-fold cross-validation was applied across all
models to enhance parameter estimation and reduce the risk of overfitting, providing a
robust and reliable model development process.

In addition to the three modeling approaches, an entropy-based weighting method was

considered to evaluate indicator importance based on the variability within the dataset. This
method was particularly effective for handling skewed distributions, as it assigns greater
weight to indicators with higher variability. By incorporating this method, we gained further
insights into the uniqueness and spread of the data, refining the overall weighting strategy
used in the development of the AIIPI.

3.3. Model Evaluation

The model performance on unseen data was assessed using Root Mean Square Error (RMSE),
R-squared (R²), and Mean Absolute Error (MAE) to compare the Elastic Net, Linear Regression,
and Random Forest models. Figure 2 summarizes the performance metrics obtained for
each model.

The Elastic Net model showed a balanced performance with an RMSE of 14.0 and an R² of
0.773, explaining 77.3% of the variance in AI investment potential. Its MAE of 2.98 indicated
moderate average prediction error. The model’s combination of L1 and L2 regularization
effectively addressed multi-collinearity, making it particularly useful for socio-economic
datasets with correlated variables. Additionally, Elastic Net’s feature selection capability
helped manage model complexity.

In contrast, the Linear Regression model had a lower R² of 0.690, indicating a weaker fit, and
a higher MAE of 3.35, suggesting greater variability in individual predictions. Despite having
a lower RMSE of 9.65, Linear Regression’s inability to capture non-linear relationships limited
its performance in this context. Its main advantage was its simplicity and ease of
interpretation.

The Random Forest model demonstrated strong performance by minimizing extreme errors,
achieving the lowest MAE of 2.41 and an R² of 0.762. With an RMSE of 10.8, it performed between
Elastic Net and Linear Regression. Random Forest’s non-linear nature allowed it to capture
complex relationships among the variables, making it particularly effective in scenarios
where reducing large errors was critical. Additionally, its ability to provide feature
importance offered valuable insights into the key factors influencing AI investment potential.

14
 Figure 2. Model Comparison by Performance Metrics

3.4. Weighting Insights from Different Models

Weights were derived based on the variable importance from each model and
subsequently normalized so that their sum equals 1, allowing them to be interpreted as
proportions or percentages, thereby enabling direct comparison across variables.

In the Elastic Net model, the highest weights were assigned to the Number of Articles
Published (45.0%) and Population (23.5%), underscoring the pivotal role of a robust research
ecosystem and market size in attracting AI investments. Other significant contributors
included the GSMA Connectivity Index (6.4%), Statistical Data Capacity (4.8%), and Political
Stability (4.0%), indicating that innovation capacity, population size, and digital infrastructure
are key factors for AI adoption.

The Random Forest model produced a different weighting distribution, with Population
(20.3%), Number of Articles Published (13.0%), and GSMA Connectivity Index (12.7%) as the top
contributors. Additionally, the model placed considerable importance on Government
Effectiveness (10.9%) and Political Stability (10.4%), highlighting the crucial role of governance
and infrastructure in creating an attractive environment for AI investments. The spread of
weights among these indicators emphasized the model's focus on governance quality and
digital readiness as foundational elements for successful AI implementation.

15
The Entropy-Based Weighting method emphasized the significance of the Number of
Articles Published (36.3%) and Population (34.5%), with AI Strategy (10.4%) also emerging as a
notable contributor. This approach reinforced the importance of a proactive national AI
strategy, a strong research base, and market size in drawing AI investments. The entropy-
based method’s reliance on variability allowed it to capture unique aspects of each
indicator, particularly highlighting those with greater diversity across countries, making it a
valuable complement to the other models by emphasizing different dimensions of
importance.

Furthermore, the correlation analysis results, reported in Figure 3, revealed strong alignment
of composite index based on the weights from Elastic Net model and the index based on the
weights from the Random Forest models, with both identifying key drivers of AI investment
potential, such as the number of AI-related articles and population size. Despite
methodological differences, these models consistently highlighted similar factors,
reinforcing the robustness of the AI Investment Potential Index (AIIPI).

16
 Figure 3. Correlation Matrix of Composite Indices

In contrast, the entropy-based weighting method provided a distinct perspective, placing

greater emphasis on population size and research output, while also giving significant
weight to AI strategies and other country-specific attributes. This approach captured
additional variability, offering a complementary view of AI investment potential, enriching
the overall analysis with insights not emphasized by the other models.

This comparative analysis of the weighting approaches illustrates how each method brings
a unique perspective, allowing for a nuanced understanding of the factors driving AI
investment decisions.

17
3.5. Variable Importance from Best Performing Model - Random Forest

In the random forest model, which was the best performing model on unseen data, the
variable importance analysis revealed several key factors influencing AI investment
attractiveness (See Figure 4; Annex 2, Table 4). Population, accounting for 20.3% of the
variable importance, emerged as a major driver, suggesting that larger populations offer
greater market expansion opportunities and workforce availability, thereby aligning with the
economies of scale necessary for AI technology adoption. The number of research articles,
contributing 13.0%, was a strong indicator of a country’s innovation capacity, correlating
directly with increased AI investments. This highlights the importance of advanced
knowledge ecosystems, as countries with significant research outputs are better positioned
for AI development and implementation. The GSMA Mobile Connectivity Index, at 12.7%,
emphasized the critical role of digital infrastructure in AI deployment, particularly in
developing economies where mobile connectivity is essential for both data collection and
operational execution of AI technologies.

Figure 4. Variable Importance Weights (Random Forest)

18
Other influential variables included Statistical Data Capacity (11.3%) and Government
Effectiveness (10.9%), both of which underscored the importance of governance and data
readiness. Countries with the ability to efficiently collect and process data, alongside robust
governmental structures, were more likely to attract AI investments. Finally, Political Stability,
comprising 10.4% of variable importance, was identified as a pivotal factor. Stable political
environments offer predictability for investors, making such countries more appealing for
long-term AI projects due to reduced risks and uncertainties in governance. These findings
underscore the multifaceted nature of AI investment decisions, where population size,
innovation capacity, digital infrastructure, data governance, and political stability
collectively shape the landscape.

Figure 5. Histogram of AIIPI by Investment Potential Stage

The analysis of AIIPI scores reveals distinct trends across investment potential stages (See
Figure 5). The stages are defined as follows: Stage 1 (AIIPI < 26), Stage 2 (AIIPI between 26 and
50), Stage 3 (AIIPI between 51 and 75), and Stage 4 (AIIPI >= 76). Stages 2 and 3 show a
clustering of scores around the median, indicating stability, whereas Stages 1 and 4 exhibit
lower frequencies and narrower distributions. The density plot (See Figure 6), centered
around a mean of 49.68, suggests a balanced dispersion of investment potentials, with most
opportunities falling within a mid-level potential range, offering a moderate risk-reward
profile.

19
Figure 6. Density Plot of AIIPI

20
4. Analysis and Findings

This section delves into a comprehensive analysis of the AI Investment Potential Index (AIIPI 4),
exploring its performance at both continental and regional levels. The AIIPI provides new
insights into the readiness and attractiveness of countries for AI investments, distinctively
positioning itself from existing indices such as the Oxford Insights Government AI Readiness
Index and the IMF AI Preparedness Index (Annex 4, Table 5). By comparing these indices, we
highlight the unique factors captured by AIIPI that enhance its predictive value. Furthermore,
this analysis identifies key patterns between AIIPI and other economic indicators, such as
income levels, offering a deeper understanding of how AI investment potential aligns with
broader development trends across regions.

4.1 Continental and Regional Insights

Map 1 provides an overview of the AI investment potential in the world while Figures 7 and 8
summarize the investment potential stages at regional and continent levels, revealing a
significant disparity in investment attractiveness (See Annex 3 on Geographical Maps of
AIIPI). Europe and North America lead with the highest scores (66.41 and 70.5, respectively),
indicating a strong capacity for attracting investment. These regions show well-established
infrastructure, economic stability, and favorable regulatory environments, placing them in
Stage 3 of investment potential.

4
The AIIPI data is publicly accessible:
[Link]

21
Map 1. AI Investment Potential Index in the World

22
 Figure 7. Investment Potential Stages by Region

In contrast, regions like Sub-Saharan Africa (35.45) and Africa as a whole (36.61) are in
Stage 2, reflecting more challenging investment climates. These areas might face barriers
such as political instability, limited infrastructure, or underdeveloped financial systems,
which make them less attractive to investors compared to their counterparts. The consistent
gap between continents and sub-regions, especially between Europe and Sub-Saharan
Africa, underscores the need for targeted policy interventions and developmental aid to
bridge the gap in investment attractiveness and attract greater economic engagement
globally.

23
 Figure 8. Investment Potential Stages by Continent

These findings highlight the critical role of regional differences in shaping investment
decisions, with distinct clusters of opportunities and challenges that demand tailored
strategic approaches for each area.

The AIIPI evaluates AI investment attractiveness across continents, revealing regional

disparities and underscoring the necessity for targeted investment strategies. Figures 9, 10,
11, 12 and 13 show the AIIPI scores obtained by countries, respectively for Oceania, Africa,
Americas, Asia and Europe.

24
● Oceania: Australia and New Zealand lead in AI investment potential, whereas countries like
Vanuatu and Papua New Guinea lag due to deficiencies in digital infrastructure. Investments
in AI literacy and telecommunications are crucial for smaller island nations. Enhancing
internet accessibility and government-led initiatives could play a key role in bridging these
gaps.

Figure 9. AI Investment Potential Index by Country in Oceania

● Africa: Morocco, Mauritius, and Gabon are leaders, while countries like Eritrea and South Sudan
exhibit significant gaps in infrastructure and digital readiness. Establishing local AI hubs,
fostering international collaborations, and enhancing workforce skills can substantially
improve AI attractiveness. Public-private partnerships are also vital in addressing infra­
structural shortcomings.

Figure 10. AI Investment Potential Index by Country in Africa

25
● Americas: The U.S. and Canada rank highest, followed by Brazil and Argentina. However,
countries like Venezuela face considerable challenges, including economic instability
and weak governance, which hinder their AI potential. Policy interventions that focus on
stabilizing governance and building foundational digital infrastructure can significantly
bolster AI readiness in these nations.

Figure 11. AI Investment Potential Index by Country in Americas

● Asia: The UAE, Singapore, and Japan top the AI readiness rankings, benefiting from robust
government initiatives, advanced technological infrastructure, and proactive AI policies.
In contrast, Afghanistan and Yemen require foundational infrastructure investments.
Regional cooperation to share technological expertise and resources could help elevate
AI investment attractiveness across less developed parts of Asia.

Figure 12. AI Investment Potential Index by Country in Asia

26
 ● Europe: Northern and Western Europe outperform Eastern Europe, reflecting higher levels
of digital readiness, government support, and research intensity. Targeted investments in
digital infrastructure, cohesive AI policies, and cross-border collaborations could help
Eastern European nations catch up with their Western counterparts.

Figure 13. AI Investment Potential Index by Country in Europe

4.2. Trends and patterns observed

AIIPI vs. Oxford Insights Government AI Readiness Index and IMF AI Preparedness Index

To contextualize the AIIPI's distinctiveness, we compared it with two established indices: the
IMF AI Preparedness Index and the Oxford Insights Governance AI Readiness Index (Annex 4,
Table 5) details the differences in definitions and methodologies. Figure 14 presents the
correlation coefficients among these indices, highlighting both overlaps and unique aspects
of the AIIPI.

A high correlation of 0.9412 is observed between the IMF AI Preparedness Index and the
Oxford Insights Governance AI Readiness Index, indicating significant overlap in their focus
on governance quality, regulatory frameworks, and institutional strength.

The AIIPI shows a strong correlation of 0.8295 with the IMF index, suggesting that countries
scoring high on the IMF index also perform well on the AIIPI. This overlap indicates that
governance quality, infrastructure, and economic conditions are contributing factors for
both indices.

Similarly, the correlation between the AIIPI and the Oxford Insights index is 0.8359. This strong
relationship suggests that effective governance, as assessed by the Oxford Insights index,
closely aligns with the conditions that promote AI investment potential.

27
Distinctiveness of the AIIPI

While the AIIPI aligns well with these indices, it uniquely focuses on the specific dynamics that
enhance a country's attractiveness for AI investments. By incorporating additional
investment-oriented, economic, and market-specific indicators, the AIIPI provides a tailored
analysis of the conditions that investors prioritize.

The slightly lower correlation coefficients – 0.8295 with the IMF and 0.8359 with Oxford
Insights – underscore the AIIPI's unique perspective. This enhanced focus on economic
performance, market potential, and direct policy incentives makes the AIIPI particularly
valuable for stakeholders seeking a deeper understanding of the AI investment landscape.
It serves as a complementary tool that enriches traditional readiness measures, offering
comprehensive insights into AI investment potential across different regions.

Figure 14. Correlation of AI Investment Potential Index (AIIPI) with Existing Indices

28
It is worth noting that countries positioned in Stages 3 and 4 (marked in blue and green,
respectively) generally exhibit higher governance and AI readiness capacities, as
highlighted by comparisons with both the Oxford Insights Government AI Readiness Index
and the IMF AI Preparedness Index (See Figures 15 and 16). These indexes provide an
important validation for the trends observed in AIIPI scores.

Figure 15. Correlation between AIIPI and Oxford Insights Index and Countries Stages

The alignment of higher AIIPI scores with these established indexes demonstrates a strong
correlation between well-developed governance frameworks and an environment
conducive to AI adoption and scalability. Stage 2 countries (in orange) appear to be in a
transitional phase, exhibiting moderate readiness but lacking the advanced governance
attributes that characterize Stage 3 and 4 nations, as captured by the Oxford and IMF
indexes. In contrast, Stage 1 countries (in red) show lower governance capacity and AI
readiness, reflecting significant barriers—consistent with the scores reported in the Oxford
and IMF indicators—to adopting advanced AI infrastructure.

29
 Figure 16. Correlation between AIIPI and the IMF Preparedness Index and Countries Stages

Correlation between AIIPI and other indicators

The positive correlation between AIIPI and GDP per capita indicates that economic wealth is
a significant determinant of AI investment potential (See Figure 17). Stage 4 countries,
characterized by high AIIPI scores, are concentrated among high GDP per capita nations,
implying that wealthier countries are better equipped to foster a conducive environment for
AI development. In contrast, Stage 1 and 2 countries with lower AIIPI are mostly in the lower
GDP range, highlighting how economic limitations can restrict both investments and the
infrastructure necessary for AI growth.

30
 Figure 17. Correlation between AIIPI and GDP per Capita and Countries Stages

Notable regional outliers, such as Singapore and Iceland, consistently appear at the upper
end of readiness and connectivity indicators, demonstrating the distinct policy frameworks
and infrastructure these countries have implemented to maintain AI competitiveness.
Conversely, nations such as Afghanistan, South Sudan, and Burundi are repeatedly
positioned at the lower end of the spectrum across different indicators, revealing systemic
challenges that hinder their AI adoption capabilities.

Furthermore, our analysis also reveals that effective governance and robust connectivity
infrastructure are essential but interdependent drivers of AI investment potential. The
clustering of nations in Stages 3 and 4, both in terms of government effectiveness and
connectivity scores, indicates that the most significant strides in AI development are made
when both elements are present and mutually reinforcing. For countries seeking to enhance
their AI investment potential, this calls for an integrated approach: strengthening
institutional quality, expanding communication networks, and improving statistical data
capabilities. By focusing on these interlinked factors, governments can foster an
environment that not only attracts AI investment but also supports sustainable
development of AI ecosystems. Countries in lower stages, especially those with weak
governance and limited connectivity, must prioritize these foundational areas, as
addressing both concurrently can significantly amplify their AI readiness. Such a
coordinated effort can reduce disparities in global AI capabilities and facilitate more
equitable participation in the technological advancements of the future.

31
5. Policy Implications

The analysis underscores the critical necessity of strategic AI investments to bridge regional
and continental disparities in AI adoption and development:

Infrastructure Development: The advancement of digital infrastructure is essential to

fostering AI adoption, particularly in regions such as Sub-Saharan Africa, parts of Asia, and
Oceania. Investments in high-speed broadband networks, mobile connectivity, and reliable
electricity access are foundational prerequisites for AI-driven initiatives. Addressing these
infrastructural deficiencies is fundamental to enabling equitable technological
development, allowing underserved regions to actively participate in the AI revolution. In
particular, enhancing last-mile connectivity, supporting public-private partnerships, and
ensuring affordable internet access are key strategies that can significantly expand the
reach of digital infrastructure. Investments in renewable energy sources to support reliable
electricity access can also create synergies with sustainability goals, making AI adoption
more environmentally friendly and resilient.

Government Policy and AI Strategies: Governments must formulate and implement

comprehensive national AI strategies that create an enabling environment for AI
development. These strategies should encompass ethical guidelines, regulatory
frameworks for responsible AI, and incentives to stimulate private-sector engagement.
Effective policy interventions must also prioritize data privacy, encourage open data
initiatives, and develop adaptive regulatory frameworks to mitigate risks while fostering
innovation. Such strategies are crucial to ensure that AI adoption contributes meaningfully
to sustainable and inclusive development. Moreover, cross-border collaborations and
regional alliances can further strengthen AI capabilities, enabling smaller nations to benefit
from shared resources and knowledge. Governments should also focus on establishing AI
regulatory sandboxes that allow controlled experimentation, helping refine AI policies while
safeguarding public interests.

Skill Development: Building AI-related skills and capabilities is imperative for sustaining AI
growth, especially in low- and middle-income countries. Policymakers should prioritize
integrating AI and data science education into national curricula at all educational levels,
alongside promoting vocational and lifelong learning programs. Establishing targeted
training initiatives to enhance AI literacy is vital for cultivating a workforce capable of
effectively engaging with AI technologies, ensuring that AI investments generate broad-
based societal benefits. Additionally, partnerships with industry leaders and academic

32
institutions can help in creating specialized AI training centers and mentorship programs,
which are crucial for bridging the skill gap. Efforts to increase gender diversity in AI-related
fields should also be a focal point to ensure that AI development benefits from a wide range
of perspectives and talents, thereby promoting more inclusive growth.

The AIIPI reveals substantial regional disparities in AI investment attractiveness, with Europe
and North America emerging as leaders, whereas regions such as Africa and parts of Asia
significantly lag. To foster equitable global AI development, collaboration among
policymakers, international organizations, and private investors is imperative. Addressing
foundational barriers—including inadequate infrastructure, insufficient policy frameworks,
and skill deficits—will be instrumental in promoting inclusive AI growth, which is essential for
achieving the Sustainable Development Goals by 2030. Engaging in multilateral initiatives
that bring together diverse stakeholders can facilitate the sharing of best practices and
accelerate progress across lagging regions.

Policymakers should focus on enhancing digital infrastructure, governance, and research

ecosystems. Given the influence of factors such as population size and research output on
AI attractiveness, policies designed to expand AI-focused education and strengthen
research initiatives can markedly enhance a country's potential for AI investments.
Expanding funding opportunities for AI research and fostering collaborations between
academia and industry are crucial components of this strategy. The AIIPI serves as a crucial
tool for identifying strategic opportunities to cultivate robust AI ecosystems globally, thereby
guiding stakeholders toward interventions that foster equitable and sustainable
development. Furthermore, targeted interventions in areas such as AI ethics, societal impact
assessments, and international regulatory harmonization can ensure that AI growth aligns
with broader human development objectives, ultimately contributing to a fair and just global
AI landscape.

Further research could explore time-varying weights for indicators, income group-specific
weights, and non-linear relationships to better understand evolving dynamics.

33
References

Acemoğlu, D.(2024) International Development Madu, N., Ezenwa, L.,

“The Simple Macroeconomics of Research Centre (2023) Adebayo, P. & Mugume, T.
AI,” Economic Policy. “Combining forces for a new (2024)
phase of AI for development: “Ethical horizons: Mapping AI
Addo, P. M., Baumann, D., Africa and beyond”, IDRC-CRDI. policy in Africa”, AfriLabs, 33p.
McMurren, J., Verhulst, S.,
Young, A., & Zahuranec, A.J. International Telecommunication Nahar, S. (2024)
(2021) “Emerging uses of Union (2023) “Modeling the effects of
technology for development: A “United Nations activities on artificial intelligence (AI)-based
new intelligence paradigm”, artificial intelligence (AI) 2023”, innovation on sustainable
AFD éditions, Policy Paper No.6. International Telecommunication development goals (SDGs):
[Link] Union. Applying a system dynamics
s/emerging-uses-technology­ perspective in a cross-country
development-new­ Kaufmann, D. & Kraay, A. (2023) setting”, Technological
intelligence-paradigm “Worldwide Governance Forecasting and Social Change,
Indicators, 2023 update”, World vol. 201, April 2024, Elsevier, 27p.
Al Yahya, D. (2023) Bank. [Link]
“Why tech diplomacy is key to 2023.123203
embracing the digital Khanal, S., Zhang, H.,
economy”, World Economic & Taeihagh, A. (2024) OECD
Forum. “Building an AI ecosystem in a “Artificial intelligence”, OECD n.d.
small nation: Lessons from
Chui, M., Hazan, E., Roberts, R., Singapore’s journey to the OECD (2024)
Singla, A., Smaje, K., forefront of AI”, Humanities and “Explanatory memorandum on
Sukharevsky, A., Yee, L. Social Sciences Communications. the updated OECD definition of
& Zemmel, R. (2023) [Link] an AI system”, OECD Artificial
“The economic potential of 03289-7 Intelligence Papers, n° 8,
generative AI: The next Editions OCDE, Paris.
productivity frontier”, McKinsey Kharas, H. (2021) [Link]
& Company, 68p. “Global development 019-14108-y
cooperation in a COVID-19
Fraisl, D., See, L., Fonteneau, F., world”, Center for Sustainable Oxford Insights (2023)
& Jütting, J. (2024) Development, Brookings “Government AI Readiness
“AI through the lens of official Institution, 18p. Index 2023,” Oxford Insights.
statistics and the Sustainable
Development Goals: The Le, T.-H., Bui, M.-T. Roszkowska, E., Filipowicz-
benefits and risks”, PARIS21, & Uddin, G.S. (2022) Chomko, M., Łyczkowska­
The Partnership in Statistics for “Economic and social impacts Hanćkowiak, A., Majewska, E.
Development in the 21st of conflict: A cross-country (2024)
Century, Discussion Paper No. 17. analysis”, Journal of “Extended Hellwig’s Method
Environmental Studies, Utilizing Entropy-Based Weights
Goldman Sachs (2023) Economic Modelling, Elsevier, and Mahalanobis Distance:
“Generative AI Could Raise 13p. Applications in Evaluating
Global GDP by 7%”, Goldman [Link] Sustainable Development in the
Sachs. .2022.105980 Education Area”, Entropy, 20p.
[Link]

34
United Nations (2024) Vinuesa, R., Azizpour, H., World Economic Forum. (2024)
“UN telecoms agency chief: Leite, I., Balaam, M., Dignum, V., “Strategic Intelligence”, The Digital
One third of humanity still offline”, Domisch, S., Felländer, A., Economy.
UN News, Global perspective Langhans, S., Tegmark, M., [Link]
Human stories. & Fuso Nerini, F. (2020)
“The role of artificial intelligence
United Nations Environment in achieving the Sustainable
Programme (2023) Development Goals”, Nature
“Climate change undermines Communications, 11, 233.
nearly all sustainable [Link]
development goals”, UNEP. 019-14108-y

United Nations Statistics Willige, A. (2024)

Division (2024) “From virtual tutors to
“The Sustainable Development accessible textbooks: 5 ways AI
Goals Report”, United Nations. can benefit education”,
[Link] The World Economic Forum.
ort/2024/The-Sustainable­ [Link]
Development-Goals-Report­ s/2024/05/ways-ai-can­
[Link] benefit-education/

35
Annex 1 Data sources and information on indicators

Table 2. Data sources and information on indicators

Description of
Indicators Primary Source Source of the source
Indicators

Government Average answer to Oxford Insights Network Readiness

Promotion of survey questions Government AI Index (Portulans
Investment in concerning the extent Readiness Index 2023 Institute)
Emerging to which different
Technologies governments foster AI Readiness Index ­ Network Readiness
investments in five Oxford Insights Index – Benchmarking
types of emerging the Future of the
technology. Network Economy

Graduates in STEM or Percentage of UNESCO

computer science graduates in STEM
from tertiary UIS Statistics
education for both ([Link])
sexes. The following
indicator has been
retrieved from a larger
database regarding
the distribution of
tertiary graduates by
field of study.

Statistical Capacity The SPI framework World Bank

assesses the maturity
and performance of Statistical
national statistical Performance
systems in five key Indicators
areas, called ([Link])
pillars. The five pillars
are: Data use, Data
Services, Data
Products, Data
Sources, Data
Infrastructure.

36
Data Privacy and Classification of data- CNIL (French Data CNIL (French Data
Protection Law related regulation by Protection Authority) Protection Authority)
[Link]
classification has
been assigned to a Data protection Data protection
percentage around the world | CNIL around the world | CNIL
corresponding to the
relative advancement
of the country in terms
of the establishment
of accountable
administrative bodies
for data protection:

• 0%: no specific
law

• 50%: partially
adequate

• 100% : data
protection
law(s)

• 100%: EU or EEA
member
country

• 100%:
Independent
authority and
law(s).

Telecommunications This index - ranging UN E-Government International

Infrastructure Index from 0 to 1- is a Knowledgebase Telecommunications
composite indicator Union
composed of four
Statistics ([Link])
indicators: the Data Center ([Link])
percentage of internet
users, the Mobile-
cellular subscriptions
per 100 inhabitants, the
active mobile-
broadband
subscriptions, the
Fixed broadband
subscriptions per 100
inhabitants.

37
Human Capital Index The HCI - ranging from
(HCI) 0 to 1 - is a weighted
World Bank
average composite of
four components [Link]
retrieved from the .org/indicator/[Link].O
UNESCO-UIS : Adult VRL
literacy (25%), Gross
enrolment ratio (25%),
Expected years of
schooling (25), Mean
years of schooling
(25%).

GSMA Connectivity The Mobile Oxford Insights GSMA Mobile

Index Connectivity Index Government AI Connectivity Index
measures the Readiness Index 2023
performance of 173 2024 – GSMA Mobile
countries against the AI Readiness Index ­ Connectivity Index
key enablers of mobile Oxford Insights
internet adoption.

Access to Electricity The percentage of World Bank World World Bank

population with Development
access to electricity. Indicators via [Link]
Electrification data are DBnomics: Access to .org/indicator/[Link].A
collected from electricity (% of [Link]
industry, national population)
surveys and
international sources. WB/WDI | DBnomics

38
Government Reflects the World Bank Worldwide World Bank Worldwide
Effectiveness perception of the Governance Indicators Governance Indicators
quality of public
services, the quality of Home | Worldwide Home | Worldwide
the civil service and Governance Indicators Governance Indicators
the degree of its ([Link]) ([Link])
independence from
political pressures, the
quality of policy
formulation and
implementation, and
the credibility of the
government's
commitment to such
policies. It ranges
from approximately ­
2.5 to 2.5.

Control of Corruption Reflects perceptions of World Bank Worldwide

the extent to which Governance Indicators
public power is
exercised for private Home | Worldwide
gain, including both Governance Indicators
petty and grand forms ([Link])
of corruption, as well
as "capture" of the
state by elites and
private interests. It
ranges from
approximately -2.5 to
2.5.

39
Rule of Law Reflects perceptions of World Bank Worldwide
the extent to which Governance Indicators
agents have
confidence in and Home | Worldwide
abide by the rules of Governance Indicators
society, and in ([Link])
particular the quality
of contract
enforcement, property
rights, the police, and
the courts, as well as
the likelihood of crime
and violence. It
ranges from
approximately -2.5 to
2.5.

Regulatory Quality Reflects perceptions of World Bank Worldwide

the ability of the Governance Indicators
government to
formulate and Home | Worldwide
implement sound Governance Indicators
policies and ([Link])
regulations that
permit and promote
private sector
development. It
ranges from
approximately -2.5 to
2.5.

Political Stability Measures perceptions World Bank Worldwide

of the likelihood of Governance Indicators
political instability
and/or politically- Home | Worldwide
motivated violence, Governance Indicators
including terrorism. It ([Link])
ranges from
approximately -2.5 to
2.5.

40
Voice and Reflects perceptions of World Bank Worldwide
Accountability the extent to which a Governance Indicators
country's citizens are
able to participate in Home | Worldwide
selecting their Governance Indicators
government, as well as ([Link])
freedom of expression,
freedom of
association, and a free
media. It ranges from
approximately -2.5 to
2.5.

GDP per Capita PPP Provides per capita World Bank World Bank
values for gross
domestic product [Link] [Link]
(GDP) expressed in .org/indicator/[Link].P .org/indicator/[Link].P
current international [Link] [Link]
dollars converted by
purchasing power
parity (PPP) conversion
factor.

GDP per Capita GDP per capita is World bank World Bank
(current $) gross domestic [Link]
product divided by GDP per capita .org/indicator/[Link].P
midyear population. (current US$) [Link]

Population This indicator is based World Bank World Bank

on the de facto
definition of [Link] [Link]
population, which .org/indicator/[Link].T .org/indicator/[Link].T
counts all residents OTL OTL
regardless of legal
status or citizenship.
The values shown are
midyear estimates.

41
Number of articles The number of AI Emerging Technology The Merged Academic
articles published by Observatory Corpus (MAC) is a
authors from the dataset that is not
country over the past Country Activity publicly available but
decade. Author Tracker: Artificial is composed of data
countries are inferred Intelligence from Clarivate’s Web
from where their of Science platform,
Country Activity
organizations are The Lens, The arXiv
Tracker: Artificial
located over a dataset platform, Papers with
Intelligence ([Link])
of more than 260 Code, Semantic
million scholarly Scholar and OpenAlex.
articles.

Percentage of articles Percentage of articles

with international per country lists
collaboration authors from
organizations located
in more than one
country over a dataset
of more than 260
million scholarly
articles.

Patents granted Patents granted CAT uses patent data

counts the number of from 1790 Analytics,
grants per country for PATSTAT, and The Lens.
over 360,000 AI-
related patent
families, which are
groups of patent
documents related to
the same invention.

Percentage of patent Growth in granted

growth patents from 2018 to
2021 shows the
evolution of granted
patents from 2018 to
2021.

42
Number of incoming Counts the number of Crunchbase
investments incoming investments (commercial
for each country datasets)
per year.
Crunchbase: Discover
innovative companies
and the people behind
them

43
Annex 2 Descriptive Statistics of Unscaled Data, and Table of Weights
from the Best Model

Table 3. Minimum and Maximum Values of unscaled data

Variable Min Max

[Link] 0 100

[Link] 32 100

[Link] 33.803 93.477

[Link] 0 100

[Link] 7.332 11.894

[Link] -1.889 2.285

[Link] 0.204 1

[Link] 0 1

[Link] -2.475 1.468

Population 47642 1417173173

[Link] -1.778 1.775

[Link] 4 575258

%.[Link] 3 132672

[Link] 34.8 93.6

Example Calculation

For the Population variable, where the minimum population is 47,642 and the maximum is
1,417,173,173, a country with a population of 41,128,771 would have a scaled value of
approximately 2.90. The scaled value is computed as:

Scaled Population = (41,128,771 − 47,642) / (1,417,173,173 − 47,642) × 100 = 2.90

44
 Table 4. Random Forest Variable Importance Weights

Variable Random Forest Variable Importance Weights

Population 0.202523637

Number of Articles 0.129985020

GSMA Connectivity Index 0.126591159

Statistical Data Capacity 0.113244356

Government Effectiveness 0.109207093

Political Stability 0.104082874

Human Capital Index 0.072933423

Log GDP per Capita PPP 0.050017260

Voice and Accountability 0.047246416

Data Privacy and Protection Score 0.023939106

Telecommunication Infrastructure Index 0.012127162

Access to Electricity Data 0.006549224

AI Strategy 0.001553270

45
Annex 3 Geographical Maps on AI Investment Potential Index

Map 2. AI Investment Potential Index in Africa

46
Map 3. AI Investment Potential Index in Latin America

47
Map 4. AI Investment Potential Index in Asia

48
Map 5. AI Investment Potential Index in Europe

49
Map 6. AI Investment Potential Index in North America

50
Map 7. AI Investment Potential Index in Oceania

51
Annex 4 Table Comparison of AI Investment Potential Index (AIIPI)
with Existing Indices

Table 5. Comparison of AIIPI, IMF AI preparedness and Oxford Insights

IMF AI Preparedness AI Investment Oxford Insights

Aspect
Index (AIPI) Potential Index (AIIPI) Government AI
Readiness Index

Definition & Measures AI Identifies global AI Evaluates 181

Objectives preparedness across investment potential governments’
174 countries, among 193 countries, readiness to adopt AI,
focusing on digital integrating AI focusing on digital
infrastructure, human readiness with socio­ capacity, governance,
capital, technological economic, and innovation
innovation, and governance, and ecosystems. Aims to
regulatory macroeconomic guide government AI
frameworks. factors to promote strategies.
Designed to offer a inclusive and
broad view of AI sustainable growth.
readiness.

Methodology Normalizes sub- Utilizes advanced Uses a composite

indicators to a 0-1 machine learning index methodology,
scale and aggregates techniques (Elastic with equal weighting
using simple Net, Random Forests) of indicators. Data is
averaging and then to identify variable normalized through
average of first importance to predict min-max scaling.
principal components AI incoming
(PCA) of each investments. The
dimension. All weights obtained
indicators are treated based on variable
equally, potentially importance. An
missing complex entropy-based
interactions. weighting is also used
to assign importance
variable indicators.
The best evaluation
model and composite
index is selected.

52
 Principal 1. Digital 1. AI Readiness: 1. Government AI
Indicators Used Infrastructure: Telecommunication Strategy: Vision, Digital
Internet access, infrastructure, human capacity, governance
secure internet capital, mobile and ethics,
servers, broadband connectivity. adaptability.
subscriptions, etc.
2. Socio-economic 2. Technology Sector:
2. Human Capital and Factors: Population, Technology sector
Labor Market Policies: Governance, political maturity, innovation
Education levels, stability and capacity, human
digital skills, STEM government capital.
graduates, internal effectiveness
labor market
3. Data and
productivity, flexibility 3. Macroeconomic Infrastructure:
of wage
Outlook: GDP, access Infrastructure, data
determination, etc. to electricity; availability, data
purchasing power representativeness.
3. Technological
Innovation and 4. Innovation
Economic Integration:
Capacity: AI research
R&D spending, AI- articles, data
related patents,
capacity, etc.
scientific publications,
mean tariff rate, Free
movement of capital
and people, etc.

4. Regulatory and
Ethical Frameworks:
Government
effectiveness, legal
adaptability.

53
List of Tables, Figures and Geographical Maps

Tables

Table 1 Descriptive Statistics of Scaled Variables

Table 2 Data sources and information on indicators

Table 3 Minimum and Maximum Values of unscaled data

Table 4 Random Forest Variable Importance Weights

Table 5 Comparison of AIIPI, IMF AI preparedness and Oxford Insights

Figures

Figure 1 Correlation Plot on Scaled data

Figure 2 Model Comparison by Performance Metrics

Figure 3 Correlation Matrix of Composite Indices

Figure 4 Variable Importance Weights (Random Forest)

Figure 5 Histogram of AIIPI by Investment Potential Stage

Figure 6 Density Plot of AIIPI

Figure 7 Investment Potential Stages by Region

Figure 8 Investment Potential Stages by Continent

Figure 9 AI Investment Potential Index by Country in Oceania

Figure 10 AI Investment Potential Index by Country in Africa

Figure 11 AI Investment Potential Index by Country in Americas

Figure 12 AI Investment Potential Index by Country in Asia

Figure 13 AI Investment Potential Index by Country in Europe

Figure 14 Correlation of AI Investment Potential Index (AIIPI) with Existing Indices

Figure 15 Correlation between AIIPI and Oxford Insights Index and Countries Stages

Figure 16 Correlation between AIIPI and the IMF Preparedness Index and Countries Stages

Figure 17 Correlation between AIIPI and GDP per Capita and Countries Stages

54
Geographical Maps

Map 1 AI Investment Potential Index in the World

Map 2 AI Investment Potential Index in Africa

Map 3 AI Investment Potential Index in Latin America

Map 4 AI Investment Potential Index in Asia

Map 5 AI Investment Potential Index in Europe

Map 6 AI Investment Potential Index in North America

Map 7 AI Investment Potential Index in Oceania

55
 Agence française
de développement
5, rue Roland Barthes
75012 Paris l France
[Link]

What is AFD?
Éditions Agence française de développement publishes
analysis and research on sustainable development
issues. Conducted with numerous partners in the Global
North and South, these publications contribute to a
better understanding of the challenges faced by our
planet and to the implementation of concerted actions
within the framework of the Sustainable Development
Goals.
With a catalogue of more than 1,000 titles and an
average of 80 new publications published every year,
Éditions Agence française de développement promotes
the dissemination of knowledge and expertise, both in
AFD’s own publications and through key partnerships.
Discover all our publications in open access at editions.
[Link].
Towards a world in common.

Publication Director Rémy Rioux

Editor-in-Chief Thomas Melonio

Legal deposit 4th quarter 2024

ISSN 2492 - 2846

Rights and permissions

Creative Commons license
Attribution - No commercialization - No modification
[Link]

Graphic design MeMo, Juliegilles, D. Cazeils

Layout Denise Perrin, AFD
Printed by the AFD reprography service

To browse our publications:

[Link]