The Global

AI Index
Methodology Report

june 2023
 Overview
Launched in 2019, the Global AI Index (GAII) was the first to rank countries based on
capacity for artificial intelligence, by measuring levels of investment, innovation and
implementation. For the fourth iteration of the index, Tortoise has worked to further
reflect the current international landscape across the areas of talent, infrastructure,
operating environment, research, development, commercial ventures and government
strategy. For the first time, this year the Global AI Index can also be explored by isolating
per-capita indicators. This allows us to understand the degree of intensity of countries’ AI
development - benchmarking growth against themselves – and expand our analysis
to small but dynamic countries, beyond the big global AI leaders.

This report details the underlying methodology of the Global AI Index, including the
rationale for its structure and the techniques behind the data collection, imputation,
weighting and scoring. As a composite index, the GAII draws on a range of 28 different
data sources, including government reports, public databases from international
organisations, think-tanks and private companies, as well as Tortoise’s own research,
to measure the national ecosystems that determine capacity for artificial intelligence.
The 111 indicators that comprise the Global AI Index Index have been selected
because they:

• Reflect publicly-available information

• Use up-to-date data sources
• Relate to key issues in the artificial intelligence sector

The indicators are grouped by associative themes around three main pillars and seven
sub-pillars:

• I mplementation. Indicators within this pillar reflect the operationalising of artificial

intelligence by practitioners in business, government and communities. This pillar
contains the sub-pillars of talent, infrastructure and operating environment.
• Innovation. Indicators within this pillar reflect technology breakthroughs and
advancements in methodology that are indicative of greater capacity for artificial
intelligence in the future. This pillar contains the sub-pillars of research and
development.
• Investment. Indicators within this pillar reflect financial and procedural commitments
to artificial intelligence. This pillar contains the sub-pillars of commercial ventures and
government strategy.

1
 Why measure AI capacity?
Artificial intelligence holds enormous power to transform business, government
and society. Measuring countries’ AI capacity means understanding the extent to which
they are set to harness such power.

Capacity – the amount of something that a system can contain or produce – is in fact
the organising concept of the Global AI Index. It is an appropriate means of considering
the relationship between the different relevant factors that exist within a given nation.
Increased capacity, in this case, can be understood as an increased ability to generate
and sustain artificial intelligence solutions, now and in the future.

At a national level, greater adoption of artificial intelligence means that more systems,
initiatives and personnel are becoming active in the field, and the quality of these factors
is also improving. In this way, capacity for artificial intelligence expresses both the
breadth and depth of adoption as well as improvements in a given nation’s ability
to manage and sustain artificial intelligence systems in a productive, safe and fair way.

Within the Global AI Index, capacity is measured through composite indicators which
meet the need to consolidate – through aggregation – a large amount of data into a set of
simplified numbers that encompass and reflect the underlying complexity of information.

Guiding principles
The key methodological principles that underpin the Global AI Index are detailed below:

1. R elevant. Each of our variables speaks to a contemporary policy area, or ongoing

conversation in business in the field of artificial intelligence. For example, ‘Number of
Paper in accepted IEEE Papers on artificial intelligence topics’ is a factor that features
regularly in contemporary discussion.
2. Relatable. Many of our variables are selected to be accessible to specialists and non-
specialists alike. This accessibility makes the Index more transparent, allowing users
to question inclusions and the relationships that they show. The phrasing of each
indicator should be clear and understandable.
3. S izable contribution. Finally, our indicators are selected due to the sizable contribution
that they make to the overall level of capacity in a given nation. In this sense, we have
aimed to include indicators that are widely referenced and considered on the basis of
their significance. For example, the ‘Number of Data Scientists/Engineers’ is widely
regarded in commentary as not only relevant and relatable as a means of measuring
some nations’ capacity, but it is also seen as making a significant contribution to
that capacity.

The Global AI Index includes mainly quantitative data (e.g. the number of data scientists,
artificial intelligence startups or GitHub commits). In a small number of cases, qualitative
data is included (e.g. IPSOS “I do not trust artificial intelligence survey” response data)
and is packaged as quantitative for comparability purposes.

2
 Update for the Global AI Index 4.0
All changes made to the methodology for this fourth edition of the Index and rankings
are detailed in this report.

Gross Capacity and Intensity

The Global AI Index has always been composed of both gross and proportional (per-
capita) indicators. For many indicators, it is therefore possible to create both a gross,
and a proportional score relative to population or national GDP. This has allowed us to
compare nations of vastly different sizes, when considering capacity. This year we have
presented the Index both as a combined score (i.e. featuring both combined, gross and
proportional indicators) and as an intensity-only score. We believe it is worthwhile to
highlight both the gross capacity for artificial intelligence according to our data, as well as
the intensity - a measure in which many smaller nations will rise to the top.

Pillars and sub-pillars

This section shows the organisation of the sub-pillars and offers a justification for
their inclusion in the Global AI Index, along with their constituent indicators. These
justifications reflect our understanding of the interrelated factors that contribute to
capacity on a national scale, knowing that the fast-changing processes of innovation and
implementation in artificial intelligence will require constant re-examination.

Implementation Talent
Artificial intelligence is implemented by people. This refers to the everyday practitioners
of artificial intelligence who are employed by the public and private sector to apply
technology to specific problems. Capacity, therefore, is based substantially on the
personnel able to deploy, manage and implement technology systems.
The geographical concentration of AI specialists, their movements, and the changing
supply and demand for them across the field is the focus of the ‘Talent’ sub-pillar. The
purpose of measuring talent is to define the level of capacity offered by human capital
within a given nation.

Implementation Infrastructure
Reliable infrastructure from basic electricity and internet access, as well as super
computing capabilities and deep databases are required to sustain the operationalisation
of different artificial intelligence solutions, and increase AI adoption.
The baseline infrastructure needed to maintain and expand artificial intelligence
initiatives includes a steady electricity supply and fast broadband speeds in cities,
smaller towns and rural areas, while more advanced infrastructure includes
supercomputing capacity.

Implementation Operating environment

Technologies thrive when the wider society approves of them. Countries should also
focus on developing strong public information initiatives, trust, and recognition in the
public sphere. Capacity results from a conducive operating environment.
The operating environment stands for political, social, legislative, economic, cultural
and natural environmental factors that significantly affect the implementation of AI
technologies. The ‘Operating Environment’ sub-pillar focuses on survey data indicating
trust in artificial intelligence, the diversity of practitioners, visa processing and data
3 governance as facilitating factors.
 Innovation Research
Research and researchers generate new ideas in artificial intelligence. Capacity as a result
of research is substantially based upon the level of activity amongst research communities
and the extent to which they share and propagate ideas.
Measuring the level of research includes an assessment of the number of papers, and the
citations, their impact according to the Computer Science H-Index, as well as attendance
at conferences and contributions to IEEE journals, among others. The ‘Research’
sub-pillar is indicative of the advances in capability that contribute to capacity through
new innovations.

Innovation Development
Innovation is demonstrated by the development of new techniques and advancements –
especially in the field of artificial intelligence.
The ‘Development’ sub-pillar focuses on collaboration on open source artificial
intelligence platforms, the ISO Artificial Intelligence Committee status, and several
indicators describing the level of patentable innovation.

Investment Commercial ventures

Commercial ventures - businesses that are providing goods and services through the
combination of financial and industrial aspects - are responsible for a large proportion
of the implementation of artificial intelligence around the world. The scale, funding and
volume of these businesses is a contributor to capacity.
The increases in productivity, efficiency and reliability that machine learning can
provide are all significant enhancements to business performance in many sectors.
The ‘Commercial Ventures’ sub-pillar is focused on the industrial environment
surrounding artificial intelligence in a given country, analysing the number, scale and
funding of ventures.

Investment Government strategy

Government strategies - often a collection of publications outlining approaches to digital
transformation, innovation and artificial intelligence – detail commitments to invest and
align interests in research communities.
The content, presentation and apparent urgency of these national strategies is the
focus of the ‘Government strategy’ sub-pillar, with indicators measuring AI spending
and targets.

4
 Geographical scope
The rapid transformation of public and private sector activities by artificial
intelligence is a global phenomenon. Whilst the pace of change is faster in some
regions than others, the Global AI Index is global in scope. With the aim of including
as many nations as possible whilst maintaining the robustness and relevance of the
underlying dataset, the fourth edition of the Index includes 62 countries. These are:
Argentina, Armenia, Australia, Austria, Bahrain, Belgium, Brazil, Canada, Chile,
China, Colombia, Czech Republic, Denmark, Egypt, Estonia, Finland, France,
Germany, Greece, Hong Kong, Hungary, India, Indonesia, Iceland, Ireland, Israel,
Italy, Japan, Kenya, Lithuania, Luxembourg, Malaysia, Malta, Mexico, Morocco,
Nigeria, Norway, New Zealand, Pakistan, Poland, Portugal, Qatar, Russia, Saudi
Arabia, Singapore, Slovakia, Slovenia, South Africa, South Korea, Spain, Sri Lanka,
Switzerland, Sweden, Taiwan, The Netherlands, Tunisia, Turkey, United Arab
Emirates, United Kingdom, United States of America, Uruguay, Vietnam.

In future editions we hope to expand the geographical scope of the Global AI Index
to include more countries. This will deepen the comparative relevance of the GAII,
and further acknowledge the role played by countries outside the global north and the
OECD in the rise of artificial intelligence solutions.

We validated the list of nations included in the Index in discussion with experts, as
well as through literature review. Beginning with countries that published stated
initiatives related to artificial intelligence provided a basis for selecting countries with
sufficient data available for ranking.

Missing values
Missing values represent approximately 5% of the collected dataset for the Global AI
Index. There was a limited amount of data available with which to train an imputation
model – although this was strongly considered as an option – and as such there are
a variety of imputation techniques employed. Each indicator has been individually
considered and one of the following strategies applied:

• Imputation by zero – used when data is not pre-defined but is the logical or
necessary value. E.g. if the number of Kaggle Grandmasters is empty it is most likely
because a country has never had one. For certain indicators where a lower score
indicated better performance (e.g. Top Rank in H-Index table of Computer Science
Academics), the “highest” (i.e. worst) score was imputed as zero would be.
• Imputation by average value – used when the variable in question is independent of
a country’s population size or GDP, placing the mean or median value in place of a
missing value.
• Imputation by last observation carried forward – used when alternative data
sources show only values from previous years. In some cases previous values are
taken as indicators of a country’s current state.
• Imputation by model – used in observation of obvious relationships between a
country’s demographics – population, GDP, employment rates, etc. In some cases
it was necessary to build a generalised linear model to predict what value should
be used.

5
 • Imputation by aggregated k-nearest neighbour - used in the several cases where
a model is inadequate, an average of all countries is not sensible or there are
no previous values for an indicator. In these cases we implemented a K-Nearest
Neighbours Regression model with k=3, in search of the most similar countries
(using national econometric as independent variables) to impute the value.

Normalisation and data processing

To ensure that the data underlying the Index rankings is comparable, it was necessary
to normalise each data point to a [0,1] scale. In the processing of this normalisation,
it was occasionally necessary to transform the distribution of values. For example,
a square root transformation is relevant if the higher end of the values for a particular
indicator are having a disproportionate impact on the distribution and therefore
the scoring.

The remaining variables were also normalised to comparable scales, but due to the
distribution of the data, it was sometimes deemed necessary to transform the value
(e.g. square/square root transformations) before normalising.

Many of the variables are treated in comparison to other values to ensure relevance
and comparability.

Currency
When dealing with currency values, we have converted all values to USD so that
they are comparable. As our values come from various time periods, we used an
appropriate average for each country.

Temporal coverage
The Global AI Index uses the most recent available values when possible. All values
carried forward are done so from no earlier than 2016, and a data-collection cut off
from 2016 onwards was also enforced.

For the purposes of our analysis, we have also considered the temporal range for a
‘Startup Company’ to be ‘foundation since 2016’.

6
 Scoring and weighting methodology
A country’s total score is made up of the weighted sum of its sub-pillar scores. The
score for a sub-pillar is composed of the normalised weight of categories (or “sub-sub-
pillars”) within the sub-pillar. For example, the Implementation pillar breaks down
into the Talent, Infrastructure and Operating Environment sub-pillars - the Talent
sub-pillar then breaks down further into the Learning Platforms, GitHub Activity and
Talent Pool categories.

This allows us to compare indicators within a given sub-pillar, such as Talent, rather
than comparing all individual indicators separately. In the final presentation of the
Index, the overall score and the score for each sub-pillar are normalised between 100
and the minimum original score. We have chosen to keep the minimum normalised
score as the minimum original score rather than normalising to 0, as giving a country
“0” in the overall or sub-pillar scores may inaccurately imply that there is no AI
capacity or activity taking place in the country at all.

Weighting for relevance

Each indicator has been considered according to its relevance to the investment,
innovation and implementation specific to artificial intelligence. Whilst we maintain
that all inclusions in the Index can be justified by this relevance, it is important to
reflect in the weightings that some factors are more closely related than others.
Our assessment of relevance is based on the apparent connections between the
indicator itself, and the overall change in artificial intelligence capacity. For example,
we consider ‘Existing Number of ‘AI Professionals’ to be a highly relevant factor in
contributing to capacity – and are therefore heavily weighted in the ‘Talent’ sub-pillar
of the Index. Whereas ‘Percentage access to electricity’ is considered a less relevant
factor, whilst still being clearly connected to capacity. It is, therefore, less heavily
weighted in the ‘Infrastructure’ sub-pillar of the Index.

Weighting for contribution

Each indicator has also been considered according to its contribution to overall
capacity through investment, innovation and implementation. Again, we maintain
that all inclusions make a contribution towards capacity in some way, but it is
important to reflect that some factors contribute more heavily. For example, we
consider ‘Total funding of AI startups’ to be a significant contributor to capacity –
and therefore weigh this indicator more heavily than others. Whereas ‘Proportion of
Total Integrated Circuits Exports’ is an indicator that highlights an important and
relevant factor, but is not as great a contribution to capacity.

Weighting for comprehensiveness

Finally, each indicator is considered on the basis of the comprehensiveness of the
source data-set from which it is drawn. Some sources are more complete than others
– alongside the process of imputation for missing values – it is also necessary to
account for the completeness of the data in the weighting system. In the case that
data availability is limited, we have reduced the relative weight of the indicator.
The degree of confidence that we have in the representativeness of the data means
that we can weigh this factor more heavily.
Pillar Weighting
Talent 15%
Infrastructure 11%
7 Operating Environment 6%
Research 26%
 Development 14%
Government strategy 4%
Commercial ventures 24%

The purpose of weighting for comprehensiveness is to ensure that the overall scoring
in the Global AI Index is, as far as possible, based upon reliable information; reducing
the relative impact of more unreliable data on the rankings.

What is the effect of the weightings?

Each layer of the weighting system for the Global AI Index adds to the accuracy,
completeness and explanatory value of the comparative rankings. It is intended to
account for the fact that contributions to capacity for artificial intelligence take many
different forms, and have varying degrees of impact at present and future levels.
We recognise, however, that our weightings are based on subjective assumptions, and
judgements applied in order to improve the coherence of the data. These subjective
judgements affect the composite scoring for each company and therefore their
position in the rankings.

Example weighting case

“Number of Data Scientists and Engineers” is gathered from LinkedIn returns. This
means that the level of comprehensiveness is questionable; mostly due to the fact that
many data engineers or equivalent may not be listed on LinkedIn or have sufficiently
identified themselves as such. The weighting for comprehensiveness is therefore set at
a 4. The contribution made by Data Scientists and Engineers to capacity for artificial
intelligence is significant as artificial intelligence is a generic and multidisciplinary
domain in many ways and so Data Engineers are able to work in collaboration
with practitioners in many different sectors. For these reasons, the weighting for
contribution is set at 5. Finally, the title ‘Data Scientist/Engineer’ is not specific to
artificial intelligence, in that many data engineers work across a range of tasks that do
not specifically involve artificial intelligence. However, Data Science & Engineering
work is essential to the development of AI concerning the preparation and analysis of
large datasets necessary for training Machine Learning and Large Language Models.
The relevancy is therefore set at 4.

8
 FAQs
Why have we built the Index?
Tortoise is fundamentally committed to data-driven news. We pursue a deep
understanding of processes to inform our journalism and our conversation with
members. We are also responding to the need amongst governments for a more
comprehensive tool for identifying these processes. The Global AI Index is part of our
investigation of artificial intelligence – recognising that it is one of the defining forces
shaping our world today.

Why is it an index then? And not just a set of presentations of data?

At Tortoise we believe in the agenda setting power of indices. Not only do they allow
for tracking of important processes through carefully selected metrics, but they also
invoke repeated comparison. In future editions, and by refining our methodology in
open discussion with our members, experts and any other interests we hope to base
policy recommendations, stories and observations about artificial intelligence on
relevant data. Comparison is key to this, highlighting factors that affect change in
some places.

We see this journalistic intent as complementary to a further set of strengths of the

index format; following a framework provided in the OECD review “Composite
Indicators – A review Michaela Saisana Group of Applied Statistics Joint Research
Centre European Commission”

– To summarise complex or multi-dimensional issues.

– To place countries’ performance at the centre of the policy arena.
– To offer a rounded assessment of countries’ performance.
– To enable judgments to be made on countries’ efficiency.
– To facilitate communication with ordinary citizens.
– To be used for benchmarking countries of best performance.
– To indicate which countries represent the priority for improvement efforts.
– To stimulate the search for better data and better analytical efforts.
– To set local priorities, and to seek out improvements along dimensions of
performance where gains are most readily secured.

How did you carefully select your metrics?

We selected our metrics through consultation with expert advisors, who helped
us build an understanding of the development of artificial intelligence. Next, we
conducted a careful investigation of available national strategies, highlighting the
common features and deriving a list of indicators.

Why have you presented an index ranking on capacity?

Capacity is the concept we selected as an organisational tool, bringing together
the many interconnected factors involved in developing and deploying artificial
intelligence. It refers to the amount of artificial intelligence-related factors in a given
nation. Capacity is also suitable given that the Index measures a range of inputs,
outputs and outcomes – this is because capacity refers both to the present and
potential level of development in the future.

9
© Tortoise Media 2023. No part of this index may be reproduced without permission. All rights reserved
 Indicators
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Coursera Data Implementation Talent The Coursera Global Skills Index score for 4 3 4 0.57
Science score machine learning
Stack Overflow Implementation Talent The total number of questions related to 5 1 5 1.14
Questions related artificial intelligence posed on Stackoverflow,
to AI by users from a given country.
Stack Overflow Implementation Talent The total number of answers to artificial 5 1 5 1.14
Answers to AI intelligence related questions on
related Questions Stackoverflow, by users from a given country.
Stack Overflow Implementation Talent The number of questions related to artificial 5 1 5 0.57
Questions related intelligence posed on Stackoverflow, by users
to AI per capita from a given country per capita
Stack Overflow Implementation Talent The number of answers to artificial 5 1 5 0.57
Answers to AI intelligence related questions on
related Questions Stackoverflow, by users from a given country
per capita per capita
Github Commits Implementation Talent The total number of all GitHub Commits 5 3 4 1.37
to public AI-related project repositories in a
given country.
Github Commits Implementation Talent The total number of all GitHub Commits 5 2 4 0.63
per capita to public AI-related project repositories in a
given country per capita
Number of IT Implementation Talent The total number of IT Graduates, in a given 1 3 2 0.45
Graduates per country per capita.
capita
Number of STEM Implementation Talent The total number of STEM Graduates, in a 2 3 2 0.52
Graduates per given country per capita
capita
Number of IT Implementation Talent The total number of IT Graduates, in a given 1 3 2 0.9
Graduates country
Number of STEM Implementation Talent The total number of STEM Graduates, in a 2 3 2 1.05
Graduates given country
Existing number Implementation Talent The total number of people currently 4 5 4 1.95
of Data Scientists describing themselves as a Data Scientist or
and Engineers Engineer in a given country
Number of Data Implementation Talent The total number of people currently 4 5 4 0.97
Scientists and describing themselves as a Data Scientist or
Engineers per Engineer in a given country per capita
capita
Existing number Implementation Talent The total number of people who describe 5 5 4 2.1
of AI professionals themselves as currently working as an
“Engineer”, “Researcher” or “Scientist”
in the fields of AI, Machine Learning,
Deep Learning, Computer Vision, Natural
Language Processing or Robotics on
Linkedin, in a given country.
Number of AI Implementation Talent The total number of people who describe 5 5 4 1.05
professionals per themselves working as an “Engineer”,
capita “Researcher” or “Scientist” in the fields of AI,
Machine Learning, Deep Learning, Computer
Vision, Natural Language Processing or
Robotics Linkedin per capita.
Number of Innovation Research The total number of full-time-equivalent jobs 4 4 3 0.84
researchers as researchers in STEM, in a given country.
Number of Innovation Research The number of full-time-equivalent jobs as 4 4 3 0.42
researchers per researchers in STEM, in a given country per
capita capita
Max Rank on Innovation Research The highest rank reached by a contributor 3 3 2 0.3
H-Index from a given country on the Computer
Science H-Index of top 1000 CS academics.

10
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Proportional R&D Innovation Research The total amount of public spending, 3 5 5 0.49
Spend proportional to the total GDP in that country,
spent annually on research and development.
Total R&D Spend Innovation Research The total amount of public spending spent 3 5 5 0.99
annually on research and development.
Number of Innovation Research The total number of Significant Machine 5 4 4 0.99
Significant Learning Systems developed within a Country
Machine Learning
Systems
Number of Innovation Research The total number of Significant Machine 5 4 4 0.49
Significant Learning Systems developed within a country
Machine Learning by capita
Systems per capita
Number of Innovation Research The total number of authors of Significant 5 4 4 0.99
Authors of Machine Learning Systems by Country
Significant
Machine Learning
Systems by
Number of Innovation Research The total number of authors of Significant 5 4 4 0.49
Authors of Machine Learning Systems by Country, per
Significant capita
Machine Learning
Systems per capita
Number of Paper Innovation Research The total number of citations received by top 2 5 4 1.83
Citations ‘CS academics’.
Number of Paper Innovation Research The total number of citations received by top 2 5 4 0.92
Citations per ‘CS academics’ per capita
capita
Number of AI Innovation Research The total number of academic articles 3 5 1 1.5
Articles published by top ‘CS academics’ in a given
country
Number of AI Innovation Research The total number of academic articles 3 5 1 0.75
Articles per capita published by top ‘CS academics’ in a given
country per capita
Number of Innovation Research The number of contributors, from a given 5 5 4 2.33
Submissions to AI country, accepted at established Artificial
Conferences Intelligence Conferences.
Number of Innovation Research The number of contributors, from a given 5 5 4 1.17
Submissions to AI country, accepted at established Artificial
Conferences Intelligence Conferences per capita
Number of AI Innovation Research The cumulative number of “AI” related 4 5 4 2.17
Related Papers research articles and papers per country.
Number of AI Innovation Research The cumulative number of “AI” related 4 5 4 1.08
Related Papers Per research articles and paper per country per
Capita capita
Number of Innovation Research The cumulative number of citations on 4 5 4 2.17
Citations on AI “AI” related research articles and papers per
Related Papers country.
Number of Innovation Research The cumulative number of citations on “AI” 4 5 4 1.08
Citations on AI related research articles and papers per capita,
Related Papers Per per country.
Capita
Number of Innovation Research The number of Universities in the Top 100 5 5 4 3.33
Universities in of the Times Higher Education Computer
Times Higher Science, in a given country.
Education Top
100 Computer
Science
Universities

11
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Number of Innovation Research The number of Universities in the Top 100 5 5 4 1.67
Universities in of the Times Higher Education Computer
Times Higher Science, in a given country per capita
Education Top
100 Computer
Science
Universities
per capita
Proportion of Implementation Operating The share of people, from a given country, 5 3 3 0.5
Population that Environment who answered: “Yes” to the question: “Do
Trusts AI you trust AI?”
Proportion of Implementation Operating The share of people who, when asked whether 5 3 3 0.5
Population who Environment “Artificial Intelligence Will Help or Harm
think AI is more People in Next 20 Years” answered “Mostly
helpful than help”
harmful
Cost of Visa Implementation Operating The cost in USD ($) of getting the fastest visa 2 4 3 1.5
Environment processing for high-skilled tech workers
Super computers Implementation Infrastructure The number of Top Supercomputers, 3 5 5 1.33
per capita” with according to Top500, in a given country.
“Number of Large
Non-Distributed
Super Computers
per capita
Super computers Implementation Infrastructure The number of Top Supercomputers, 3 5 5 0.67
according to Top500, in a given country per
capita
Total flops of Implementation Infrastructure The total combined Petaflops of the Top 3 5 5 1.33
Large Non- Supercomputers, according to Top500, in a
Distributed Super given country.
Computers per
Capita
Total flops of Implementation Infrastructure The total combined Petaflops of Top 3 5 5 0.67
Large Non- Supercomputers, according to Top500, in a
Distributed Super given country by capita
Computers per
Capita
Total Integrated Implementation Infrastructure Total global imports of Integrated Circuits by 3 2 4 0.38
Circuits Imports country.
Total Integrated Implementation Infrastructure Total global imports of Integrated Circuits by 3 2 4 0.19
Circuits Imports country proportional to GDP.
proportional to
GDP
Proportion of Implementation Infrastructure Total global exports of Integrated Circuits by 1 2 4 0.29
Total Integrated country.
Circuits Exports
Proportion of Implementation Infrastructure Total global imports of Integrated Circuits by 1 2 4 0.15
Total Integrated country proportional to GDP
Circuits Exports
proportional to
GDP
Average Implementation Infrastructure The average monthly download speed by 3 5 4 1.41
Download Speed country.
Proportion of Implementation Infrastructure The proportion of the total population with 5 5 5 1.76
Population using access to the internet.
Internet
Proportion Implementation Infrastructure The proportion of the total population with 5 5 5 1.76
of Population access to electricity.
with Access to
Electricity

12
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Mobile Implementation Infrastructure The total number of mobile subscriptions per 2 3 4 1.06
Penetration Rate 100 people in a given country.
per 100 persons
Dedicated Investment Government The amount of dedicated investment on 5 5 5 0.52
Spending Strategy artificial intelligence by government in a given
on Artificial nation.
Intelligence
Dedicated Investment Government The amount of dedicated investment on 5 5 3 0.22
Spending Strategy artificial intelligence by government in a given
on Artificial country, proportional to that country’s GDP.
Intelligence
proportional to
GDP
Spend Period Investment Government The length of time, in years, that is identified 5 5 5 0.26
of Dedicated AI Strategy as a specific spending window in which the
Budgets government will commit funding to building
capacity for artificial intelligence.
Time Scale of Investment Government The total length of time, in years, of the 5 2 4 0.25
Dedicated Strategy Strategy national strategy for a given country.
Government has Investment Government A categorical indicator of whether a country 5 3 5 0.3
Dedicated AI Strategy has drafted or published a specific, dedicated
Strategy artificial intelligence strategy report.
Government has Investment Government A binary indicator of whether a country 2 2 5 0.21
Dedicated AI Strategy has a specific ministry, and ministerial
Minister head, dedicated to implementing ‘artificial
intelligence’ in the public sector and beyond.
In this case, a ‘Dedicated AI Minister’ is
identified as distinct from other Digital, and
Technology Ministers.
Government has Investment Government A binary indicator showing whether or not a 5 2 5 0.28
Dedicated AI Strategy country's government has a dedicated body,
governmental or working group, on addressing the national-
body level requirements for building capacity in
artificial intelligence.
Government Investment Government A binary indicator of whether a country has 5 5 5 0.35
has Publically Strategy committed specific funding towards building
Dedicated Money capacity in artificial intelligence.
to AI
Government has Investment Government A binary indicator of whether a country 5 3 5 0.3
Measurable AI Strategy has measurable Targets and KPI’s for the
Targets advancement of their artificial intelligence
strategies.
Tracking of Investment Government A binary indicator on whether or not a 5 3 5 0.3
Previous Years Strategy country’s government has created tracking
efforts on AI mechanisms to keep abreast of progress on
their artificial intelligence strategy.
Dedicated Strategy Investment Government A binary indicator showing whether or 5 4 5 0.32
received External Strategy not a country’s government strategy in
Consultation ‘Artificial Intelligence’ has received external
consultation from academic, business
executives or other specialists external to
government.
Dedicated Strategy Investment Government A binary indicator of whether a country’s 5 5 5 0.35
mentions Training Strategy National AI Strategy mentions training or
or upskilling upskilling of the public through dedicated
initiatives.
Dedicated Strategy Investment Government This indicator shows whether the national AI 5 5 5 0.35
was Signed by Strategy strategy in a given country has received sign-
Senior Member of off from a senior member of government. 0
Government for no signoff, 1 for signoff from a minister,
and 2 for signoff from a head of state or
equivalent.

13
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Commits on Innovation Development The total number of commits to the most 5 4 4 2.67
High-Popularity widely developed open source AI packages
Open Source AI from Github
Packages
Commits on Innovation Development The total number of commits to the most 5 4 4 1.33
High-Popularity widey developed open source AI packages
Open Source from Github per capita.
AI Packages per
capita
Number of Filed Innovation Development Number of patents relating to AI filed by 5 2 3 1.37
AI Patents by applicants from a given country since 2017.
applicant
Number of Filed Innovation Development Number of patents relating to AI filed by 5 3 3 1.51
AI Patents by inventors from a given country since 2017
inventor
Number of Innovation Development Number of patents relating to AI granted to 5 4 4 1.78
Granted AI applicants since 2017.
Patents by
applicant
Number of Innovation Development Number of patents relating to AI granted, by 5 5 4 1.92
Granted AI inventor since 2017.
Patents by
inventor
Number of Filed Innovation Development Number of filed patents relating to AI filed by 5 2 4 0.75
‘AI Patents’ by applicants from each countryper capita.
applicant per
capita
Number of Filed Innovation Development Number of patents relating to AI filed by 5 3 4 0.82
‘AI Patents’ by inventors from each country per capita.
inventor per capita
Granted patents Innovation Development Number of patents relating to AI granted to 5 4 4 0.89
by applicants per applicants from each country as a share of the
capita population.
Granted patents Innovation Development Number of patents relating to AI granted to 5 5 4 0.96
by inventors per inventors from each country as a share of the
capita population.
Businesses Investment Commercial Percentage of Businesses reporting the use of 5 5 3 1.6
using Artificial AI in their business operations
Intelligence
Total Funding of Investment Commercial The total sum of all funding rounds to 5 5 3 3.19
AI Companies ‘AI companies’.
Total Funding of Investment Commercial The average amount of funding per ‘AI 5 3 3 1.35
AI Companies company’ in a given country, proportional to
proportional to GDP in that country.
GDP
Number of AI Investment Commercial The estimated total number of companies 5 5 3 3.19
Companies in a given country implementing artificial
intelligence as a key part of their business
process.
Number of AI Investment Commercial The estimated total number of companies 5 3 3 1.35
Companies per in a given country implementing artificial
capita intelligence as a key part of their business
process, per capita.
Average Funding Investment Commercial The average amount of funding per ‘AI 3 1 3 0.86
of AI Company company’ in a given country.
Number of Listed Investment Commercial The total number of ‘AI companies’ from 5 2 3 1.23
AI Companies a given country that are listed on a stock
exchange, either in that country or elsewhere
in the world.

14
Indicator Name Pillar Sub Pillar Short descriptions Relevance Contribution Reliability Overall
Weight
Number of AI Investment Commercial The total number of Initial Public Offerings 5 2 3 1.23
Companies on (IPOs) of ‘AI companies’ on a country’s stock
Country’s Stock exchange.
Exchange
Total Funding of Investment Commercial The sum of all funding rounds to ‘AI startups’ 5 5 3 2.6
AI Startups in a given country.
Total Funding Investment Commercial The sum of all funding rounds to ‘AI startups’ 5 5 3 1.3
of AI Startups in a given country proportional to GDP
proportional to
GDP
Number of AI Investment Commercial The estimated total number of ‘AI startups’ in 5 5 3 2.6
Startups a given country.
Number of AI Investment Commercial The estimated total number of ‘AI startups’ in 5 3 3 1.1
Startups per capita a given country per capita
Average Startup Investment Commercial The average amount of funding per ‘AI 5 3 3 1.1
Funding startup’ in a given country.
Number of AI Investment Commercial The total number of companies valued at over 5 4 4 1.3
Unicorns 1 Billion (USD $) in a given country.
Gender Diversity Implementation Operating The share of Science graduates, from a given 3 3 3 0.5
of Science Environment country, that identify as female.
Graduates
Gender Diversity Implementation Operating The share of IT graduates, from a given 3 3 3 0.5
of IT Graduates Environment country, that identify as female.
Level of Data Implementation Operating The level of data privacy legislation in a 3 5 5 0.58
Protection Environment country including GDPR equivalent, full
Regulation legislation, draft legislation or no legislation.
Presence of Right Implementation Operating Whether a country has agreed to a stipulation 3 2 2 0.31
to Explanation Environment of the right to an explanation of decisions
made predominantly using artificial
intelligence.
Open Data Implementation Operating A binary indicator of whether a country 3 3 4 0.44
Charter Environment has signed The International Open Data
Charter, or is a G20 country with the same
responsibilities.
Kapersky Implementation Operating The score for a given country on the 2 2 3 0.5
Cybersecurity Environment Cybersecurity Index, based largely on data
Score from cybersecurity company Kaspersky.
Level of Implementation Operating The status of a given country as to whether 5 5 5 0.67
Participation Environment it is participating, observing or not involved
of ISO AI with the ISO’s Standards Committee
Committee Dedicated to Artificial Intelligence.

In addition, we used 16 econometric indicators taken from the World Bank, Unesco and the Economist
Group in our modelling for imputing missing values.

15