in collaboration with

REPORT

AI: A Roadmap
for Governments
To Inspire
and Enable
The Next Generation
of Governments
The World Governments Summit is a global platform dedicated to shaping
the future of governments worldwide. Each year, the Summit sets the agenda
for the next generation of governments with a focus on how they can harness
innovation and technology to solve universal challenges facing humanity.

The World Governments Summit is a knowledge exchange center at

the intersection of government, futurism, technology, and innovation.
It functions as a thought leadership platform and networking hub for
policymakers, experts and pioneers in human development.

The Summit is a gateway to the future as it functions as the stage for

analysis of future trends, concerns, and opportunities facing humanity.
It is also an arena to showcase innovations, best practice, and smart
solutions to inspire creativity to tackle these future challenges.

2 World Governments Summit

3
Table of Contents

Introduction 06

Why Building An AI Roadmap Is So Urgent 08

The AI Opportunity 14

The AI Risks 18

Balancing Risk And Reward: The Role Of Governments 24

Conclusion 36
 Introduction

Artificial intelligence (AI) is experiencing

a period of explosive growth. However, this
transformative technology also presents
significant challenges, raising concerns
about job displacement, bias, privacy, and
environment impact.

This report explores the current landscape of

AI, examining both its immense potential and
the critical risks that need to be addressed. We
delve into the role of governments in managing
this powerful technology, focusing on strategies
for harnessing its benefits while mitigating its
downsides. A holistic approach with effective
policies and regulations is crucial to ensure AI
becomes a force for positive change, driving
economic growth, social progress, and a more
sustainable future.

6
7
 Section 1

Why Building
An AI Roadmap
Is So Urgent

8
AI is at an inflection point.
Today, more than ever, the technology
has the capacity to transform nearly
every aspect of life. This immense
potential is why governments
worldwide are developing and debating
comprehensive strategies to guide the
responsible deployment of AI tools that
protect both businesses and consumers
while fostering innovation.

9
While it may feel sudden, the rise of AI has GPUs have since been utilized for the rapid
been occurring for decades, backed by a mathematical calculations that are demanded
dramatic increase in computing power. For by AI. Condor Galaxy 1 uses more than 72,000
instance, in November 2023, UAE-based G42 of these GPUs. While China and the US lead
and US-based Cerebras completed the Condor the world in supercomputing capacity, not all
Galaxy 1 — a supercomputer four times more governments have the same resources to deploy
powerful than the previous best.1 There will such compute resources. Initiatives such as
be nine such supercomputers combined into the EU’s Joint Supercomputing Initiative help
an unprecedented single network by the end alleviate such challenges, along with cloud-
of 2024. Supercomputers are made possible based supercomputing, although the latter is
by graphics processing units (GPUs), originally generally not used by governments given data
designed for the video gaming industry to security concerns.
improve computer graphics.

Figure 1
Computational capacity of the fastest supercomputers in teraflops
1.1 billion
0.9 billion
0.7 billion
0.5 billion
0.3 billion
0.1 billion
0
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025
Source: Dongarra et al. (2023) — with minor processing by Our World in Data

Figure 2
Evolution of storage and memory cost
This data is expressed in US$ per Terabyte (TB). It is adjusted for inflation

100 trillion

1 trillion

100 billion

100 million

1 million

10,000
Memory
100 Flash
Solid state
Disk
1956 1970 1980 1990 2000 2010 2022
Note: For each year, the time series shows the cheapest historical price recorded until that year.
Source: John C. McCallum (2023); U.S. Bureau of Labor Statistics (2024) — with minor processing by Our World in Data

Other contributing factors to AI’s growth include accessible, while elastic cloud computing allows
the low cost of storage and computing power. for flexibility in demand when it comes to
Cloud storage is now ubiquitous and easily computer processing, memory, and storage.

10
Additionally, advancements in the mathematical with numerical solutions and have applied the
underpinnings of AI, especially machine power of the elastic cloud to solve these problems.
learning, or statistical learning, are pushing the Consumer-facing companies like Amazon,
technology forward. Researchers have developed Facebook, Google, and LinkedIn are using these
mathematical techniques to convert complex, advanced algorithms to deliver personalized
non-linear problems into linear formulations experiences to their customers.

Figure 3
Research in AI
Count of publications, 1980-2021
600

400

200

0
1980 1985 1990 1995 2000 2005 2010 2015 2020

Machine learning Computer Vision Artificial neutral network NLP semantics Fuzzy logic Search algorithms

Source: White House, The Impact of AI on the Future of Workforces in the EU and USA, 2022

And now companies of all types are funneling This allows organizations to more easily and
billions of dollars into generative AI, a sub-area quickly utilize large amounts of unlabeled
of AI that enables users to easily and quickly data to create foundation models. GPT-3 and
generate new content based on a variety of user Stable Diffusion are examples of foundational
prompts or inputs. The content generated by models, which empower users to harness the
these models includes text, images, sounds, power of language for tasks such as generating
animation, and 3D models (Figure 4). essays from text prompts (ChatGPT) or creating
photorealistic images from textual descriptions
Generative AI models identify patterns and (Stable Diffusion and ChatGPT’s Sora).2
structures within existing data using neural Custom-built generative AI foundation models
networks to generate new content. They can enable a variety of other smart solutions, as
leverage different learning approaches, shown in (Figure 5).
including unsupervised or semi-supervised
learning for training.

11
Figure 4
Overview of generative AI

Artificial
Intelligence Text generation
Generation Classification Summarization
Machine
Translation Research Content Editing
Learning

Deep Image generation

Learning Generation Editing

Generative AI Video generation

Generation Editing Game Assets

Large
language Task generation fine-tuning
models Generation Editing Game Assets

Figure 5
Generative AI use cases

1 Environment
• Published report summarization and analysis
of latest risks
• Satellite imagery analysis and generation

2 1 2 Healthcare
• Medical AI generated content and recommendations
based on patients’ needs and situation
• Fully automated drug research and development
Enabled by
Gen AI 3 Sports
Foundation • Generate different team and sport strategies
Model based on performance and opponents
3 5 • Personalized sports commentary

4 Transport
• Custom driver messaging generation based on traffic
analysis and driving behavior
4 • Summarization of traffic reports for faster decision making

5 Logistics
• Generate warehouse layout design and storage
strategies by analyzing available space
Source: CB Insights; Oliver Wyman analysis

12
One of the key breakthroughs of generative AI
is that it exhibits the three key characteristics of
a general-purpose technology — it is pervasive,
it allows for rapid improvement, and it leads to
further innovation. These technologies are powerful
enough to significantly impact productivity and
economic growth across a wide range of industries.
Unlike previous general-purpose technologies such
as the steam engine or electricity, generative AI’s
impact is expected to be much faster due to its
relative ease of diffusion and adoption.

I think we are seeing the

most disruptive force
in history; we will have
something that for the
first time is smarter than
the smartest human.
There will come a point
where no job is needed
Elon Musk in conversation
with Rishi Sunak3

13
 Section 2

The AI
Opportunity

14
AI’s capability to add value is playing
out across multiple industries. The
technology can create new business
models and revenue streams, while
also reinventing existing processes
and operations to achieve efficiency
and excellence.

AI is one of the most

important things humanity
is working on. It is more
profound than, I don’t
know, electricity or fire
Sundar Pichai4

15
At Shell, for instance, deploying AI-based its impact on healthcare through medical
predictive maintenance for more than “500,000 diagnostics and its role in professional and
refinery valves globally is estimated to yield support services, like generative AI boosting the
several hundred million dollars a year in reduced productivity of call center agents and software
maintenance costs and increased operational developers. Indeed, due to its extensive
efficiency.”5 Shell plans to deploy several AI application across various industries to improve
applications across its upstream, midstream, and performance and transform business models,
downstream operations globally with potential AI is commonly referred to as a General Purpose
savings reaching billions of dollars a year. Technology (GPT).

Quotient, AI by Oliver Wyman, recently collaborated ChatGPT Creates AI Frenzy

with energy giant to further illustrate the power of While businesses and researchers have been
AI. By incorporating advanced analytics, AI, and tapping into AI for years, the technology was
machine learning tools, Aramco navigated market thrust into the public consciousness in late
volatility and outperformed competitors by a 2022 with the launch of Open AI’s ChatGPT.
significant margin — US$ 2 per barrel.6 This large language model (LLM) may have
been a novelty for most consumers, but its
Recent case studies consistently demonstrate sudden ascendence also alerted organizations
AI’s capacity to enhance productivity. Numerous and governments to the possibilities of the
examples illustrate AI’s contributions to technology — both positive and negative. Indeed,
scientific advancements—such as its use in the frenzy for all-things AI saw companies that
controlling plasma in fusion reactors and mentioned AI in their earnings outperform
understanding protein folding—as well as those that did not on the stock market.7

Figure 6
AI hype in Q1 2023 earning

Alphabet 7 65

Meta 21 56

Microsoft 8 53

Amazon 12

Apple 4

Note: AI mentions during Q1 2022 and Q1 2023 earning calls of Big Tech companies
Source: Richter, F. (2023). Tech Giants Were All About AI This Earnings Season. Statista.

Since then, several similar LLMs have emerged, realm with need for technical knowledge to a
including Google’s Gemini, Baidu’s Ernie, and “human” realm due to natural language processing
upgraded versions of ChatGPT from Microsoft- is a watershed moment in the growth of AI.
backed OpenAI. The shift of AI from a “developer”

16
Figure 7
Global generative AI investments
US$ Billion, 2016 to 2023
25.8 25.8

Model Makers are technology companies

focusing on development, training and
deployment of generative AI models
19.0
(76%)

Applications developed that are derived

from generative AI foundation models
or including generative AI features
6.4
3.0
(12%)
2.0 (8%) 1.0 (4%)
0.8 (3%)

2022 2023 Total Model Apps Apps with Computing Operation

makers proprietary
Investment in generative AI models

Example of generative AI model makers

Model Entity Focus area

Context awareness, coherent language

GPT-4 OpenAI
understanding and generation

Specialized proficiency in a particular domain or task,

Llama 2 Meta
showcasing superior performance in targeted applications

Exceptional multi-modal capabilities, seamlessly

Gemini Google
integrating and processing information from various data

Technology High-speed language processing, geared towards

Falcon 180B
Innovation Institute efficient real-time interactions and applications

Saudi Authority for Data Available by invite for those interested in tech, culture,
ALLaM and Artificial Intelligence history, and health
(SDAIA)

King Abdullah University Specific methodologies to address the challenges

AceGPT of Science and Technology of Arabic syntax and ensure alignment with local
(KAUST) cultural values

Source: Dealroom.co, Generative AI, 2024

17
 Section 3

The AI Risks

18
Despite its potential, AI presents
real and very significant risks that
can offset all its benefits and more.
Many leaders consider these risks
as substantial to humanity’s shared
prosperity, and even existential in
the longer-term.

One thing that’s clear from

everything that has been
written so far about the risks
of AI — and a lot has been
written — is that no one has
all the answers. Another
thing that’s clear to me is
that the future of AI is not as
grim as some people think
or as rosy as others think
Bill Gates8

19
AI’s potential to impact all key areas of strategic There is also AI bias in healthcare, which
national importance has spurred action from contributes to existing health disparities for
government leaders around the world, ranging certain populations based on race, ethnicity,
from publishing a national strategy and action gender, age, or other demographic factors.
plan to creating an AI coordinating body to
monitor activity. Countries are also issuing Furthermore, an increase in inequality and
research grants and engaging in public-private insecurity could drive further social instability
partnerships, according to the AI Policy and political extremism. So-called deepfakes,
Observatory database with more than 1,000 photos and videos generated by AI, can also
policy initiatives from 69 countries maintained by spread disinformation and unrest. These could
the Organisation for Economic Co-operation and make it even harder for members of the public
Development (OECD).9 to distinguish fact from fiction. The World
Economic Forum’s Global Risk Report 2024
A primary concern for governments and society at highlights AI-generated misinformation and
large is the impact on jobs.10 While there is little disinformation as the second most severe global
doubt that AI will create new jobs and demand new risk anticipated over the next two years.
job training, job losses are inevitable. Goldman
Sachs estimates that as many as 300 million jobs There is clearly a danger of weakened democracies
could be affected by the latest wave of AI.11 from compromised privacy and erosion of trust
in governments. Without proper guardrails, AI
Even if AI’s overall impact on the labor market may be the biggest threat to privacy and real
ends up being positive in the long-term, democracy that has ever developed.
economic hardship and unemployment will likely
increase in the transition phase. MIT economist This is not speculation. There have already been
Daron Acemoglu argues that unless there is many AI-related events that have eroded trust
a shift in the paradigm of AI development — in governments and large corporations. This
including the building of AI tools that are useful includes data breaches of medical and sensitive
to human workers rather than tools that will information, deepfakes convincingly spreading
replace them — the overall net impact on the misinformation, and failures of government AI
labor market and society as whole may not be systems (an example of this were the incorrect
positive. His is not a lone voice.12 arrests and jailings in Detroit in the USA due to
false facial recognition matches).15
Questions also arise regarding copyright
infringement, with original work from publishers, AI is profoundly transforming cybersecurity by
artists, and even coders, being used without introducing highly sophisticated algorithms
consent, leading to legal action against AI giants.13 capable of breaching secure systems.
Furthermore, AI could empower individuals with
AI bias is another significant issue, perpetuating limited technical expertise to launch sophisticated
and amplifying existing biases present in the cyberattacks, potentially amplifying the frequency
data on which it is trained. For instance, racial and severity of these threats. The potential
discrimination of facial recognition technology has gravity of these risks, coupled with our limited
been studied extensively and images produced comprehension of them, is widely acknowledged
by generative AI evidently amplify existing by leaders in academia, industry, and government.
stereotypes. A study published by the National Even renowned tech pioneers such as Bill Gates,
Institute of Standards and Technology, indicated, Elon Musk, Steve Wozniak and Geoffrey Hinton,
in one-to-one matching, higher rates of false the so-called godfather of AI, have expressed
positives for Asian and African American faces concerns regarding the advancement and
relative to images of Caucasians. The differentials implementation of AI, advocating for regulatory
ranged between 10 to 100 times.14 measures to address these security threats.

20
Figure 8
Number of AI incidents and controversies
The number of newly reported AI incidents and controversies in
the AIAAIC database was 26 times greater in 2021 than in 2012
260

218

123

82
59
37

10 10 14
6

2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Source: AIAAIC Repository (2022) via AI Index (2023) — processed by Our World in Data

Specifically related to generative AI and their risks associated with data sovereignty, cultural
foundation models, governments are seeking local independence and large money outflows due to
custom-built foundation models to mitigate wide adoption of generative AI (Figure 9).

Figure 9
Risks governments seek to avoid with local custom-built generative AI foundational models

Data sovereignty Financial impact Cultural independence

• Control and manage data • Avoid large money outflows from • Protect the culture in which
within national borders wide adoption of generative AI the models will be used
• Respect privacy laws and ensure • Monetizing AI services and • Control narrative of output to
data handling complies with enable new products & services align with cultural priorities
local regulations by providing APIs to connect to • Avoid bias towards specific
• Less reliance on giant foundation models culture or set of norms
model-makers and the data • Integrating foundation models
they collect and process to enhance existing products

Example: Manage medical data Example: Charge $0.01 for each Example: Reply only appropriate
internally and avoid it going to API query to foundation model answers to a government
3rd party providers with an average of 10 million customer support service
queries, 1 per day for ChatGPT
Source: Oliver Wyman analysis

21
Beyond the direct social and national risks,
AI’s impact on climate cannot be ignored. The
necessary infrastructure behind AI, including data
centers and computing power have a growing When you think about the
carbon footprint. The Stanford AI Index Report
2023 estimated that the training run of ChatGPT-3
potential for scamming
emitted around 28 times more carbon than the people. If I was interested
average American uses in one year, and more than
90 times more carbon than the average human in investing in scamming,
uses in one year.16
it’s going to be the growth
Of course, the counter argument is that AI
can help reduce carbon emissions by — for
industry of all time and it’s
example — streamlining renewable energy enabled, in a way, by AI
storage or modeling designs to create optimal
conditions for low-energy usage. Warren Buffett17

Figure 10
CO2 equivalent emissions (tons) by selected machine learning models and real-life examples, 2022

GTP-3 (175B) 502

Gopher (280B) 352

OPT (175B) 70

Car,Avg. Incl. Fuel, 1 Lifetime 63

Bloom (176B) 25

American Life, Avg., 1 Year 18

Human Life, Avg., 1 Year 6

Air Travel, 1 Passenger, NY-SF 1

Source: Stanford University. “Machine learning (ML) platform emissions in CO2 equivalent in 2022, in tonnes.” April 15, 2023. Statista

22
23
 Section 4

Balancing Risk
And Reward:
The Role Of
Governments

24
Harnessing the benefits of AI while
mitigating the threats it poses will
not stem automatically from the
technology itself and its current
form of siloed development. It will
necessitate a concerted effort to
implement adjustments and controls
across institutions, regulations, and
technology. Governments are best
positioned to be in the driving seat
of this coordinated effort.

25
Proactive measures are required to drive Given the need for authority, regulation, and
institutional, regulatory, and technological oversight, governments are best positioned to be
adaptations and safeguards of AI. Comprehensive in the driving seat of the effort to ensure that the
AI rules have been proposed for the European development and deployment of AI is grounded in
Union’s 27 member countries with an Artificial strong ethical and safety principles that mitigate
Intelligence Act (EU AI Act). The legislation builds negative threats.
on earlier guardrails and was deemed necessary
due to the boom in generative AI as well as The public sector, however, cannot do this alone.
foundation models that underpin services such Workers who are familiar with the technology,
as ChatGPT. At the same time, many individual policy-makers, researchers who understand
governments have come forward with AI theoretical potential, and tech-moguls who closely
regulatory proposals. These often depend on the understand the possibilities of the technology’s
key economic sectors of individuals states — and real-world applications must all be part of a
some states have come forward with regulatory coordinated effort.
proposals focused on the AI use-cases most
pertinent to their key sectors (instead of general or
all-encompassing regulatory frameworks).

What Should Governments Do?

Governments and policy-makers need to adopt While there is some common ground, each
strategies that support the benefits of AI while country’s AI aspirations and strategies differ
also mitigating potential risks. To be effective, such based on factors such as their economic structure,
a strategy must be comprehensive, addressing social fabric, labor market and access to a skilled
all ecosystem elements and reevaluating the workforce. For example, a country with specific
paradigm of digital inclusion to encompass AI global champion sectors, such as Saudi Arabia
inclusion as well. with its oil and petrochemical industry or Taiwan,
South Korea, and the US for semiconductors, may
Aspirations prioritize AI applications that further improve
It is essential that nations clearly define their the performance and competitiveness of these.
national aspirations for the use of AI, outlining the On the other hand, a country prone to social bias
specific areas of focus and the desired outcomes for historical reasons, or one that is experiencing
of this transformative technology. As such, a sudden increase in immigration for instance,
governments across the globe are formulating would want to place particular focus on addressing
comprehensive goals and strategies to harness the and mitigating bias in AI systems.
capabilities of AI effectively and efficiently.
Analyzing more than a dozen national AI strategies
reveals a wide range of priorities regarding goals,
focus areas, and sectors:

26
Figure 11
AI ambitions of selected countries

General AI Leadership / Global Hub

• Elevate China to the forefront of AI innovation by achieving world-leading levels in AI

theories, technologies, and applications
• Establish China as the primary global AI innovation center
• Establish Australia as a global leader in developing and adopting trusted, secure and
responsible AI

• Compete on the international scene as a leading economy utilizing and exporting Data & AI
• Position Saudi Arabia as the global hub where the best of Data & AI is made reality
• Transform the United Arab Emirates into a world leader in AI by investing in people and
industries that are key to the country’s success

• Leverage AI to increase resilience, productivity, growth, and innovation across sectors to

continue the country’s leadership as a science and AI superpower

• Continued American leadership in AI is of paramount importance to maintaining the economic

and national security of the United States

More Focused On Specific AI Areas

• Build deep talent pools and support a national research community on AI

• Develop global thought leadership on the economic, ethical, policy and legal implications
of advances in AI

• Enhance AI research ecosystem and develop and reinforce excellence in AI educational programs
• Develop competences and skills
• Continued support for AI R&D
• Introduce AI solutions in public sector to improve e-services and the efficiency of the state
• Develop a common ethical and human-centered basis for AI
• Help public sector adopt AI to offer world-class public services
• Position Austria as a research and innovation hub for AI in key areas
• Deploy AI towards the common good
• Maintain Austria’s competitiveness
• Establish world-class AI infrastructure
• Enhance innovation capacity
• Friendly regulations

Source: Countries’ AI strategies, Oliver Wyman analysis

27
Figure 12
Focus sectors of AI strategies across countries

Target sectors

Aerospace ✓ ✓ ✓ ✓
Agriculture, Food and Forestry ✓ ✓ ✓ ✓ ✓ ✓
Arts, Media, Cultural and ✓ ✓ ✓ ✓
Creative Industries

Defence and Security ✓ ✓ ✓ ✓ ✓ ✓

Digital Economy ✓ ✓ ✓ ✓
Education ✓ ✓ ✓ ✓ ✓ ✓
Energy Resources and Utilities ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Environment ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Finance ✓ ✓ ✓ ✓
Healthcare ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Innovation ✓ ✓ ✓
Manufacturing ✓ ✓ ✓ ✓ ✓
Mobility, Logistics and ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transportation

Public Administration ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
and Governance

Research and Development ✓ ✓ ✓ ✓ ✓

Seas and Oceans ✓ ✓ ✓
Smart Cities/Construction ✓ ✓ ✓ ✓ ✓ ✓
Social and Welfare Issues ✓ ✓ ✓
Telecommunications and IT ✓ ✓ ✓ ✓
Tourism ✓ ✓ ✓
Source: Countries’ AI strategies, Oliver Wyman analysis

28
Policies and Regulations Restructuring large technology companies in
As governments develop a framework for specific areas to promote competition and
regulating AI, they should address critical innovation, while mitigating their dominance,
aspects, including ethics, potential bias in AI, is also under consideration by lawmakers.
and data privacy and security. They should also Additionally, rigorous testing should be mandatory
work to build policies to protect jobs, promote for new AI models before release. Several countries
social equity, and ensure shared prosperity. including the United States (US) and the United
Kingdom (UK) have already taken steps, with the
One approach involves tax policies that latter establishing an AI Safety Institute.
incentivize job creation to mitigate job loss
that comes with automation. This could include Regulation is not limited to tech giants; academia
lowering taxes on wages while increasing taxes needs scrutiny as well. Students using ChatGPT to
on AI software. Additionally, tax credits and write essays highlight the challenges. While AI is
targeted subsidies can be used to promote valuable for research in these institutions, students
the development of AI that complements and must develop independent skills in the long run
enhances human work, rather than replacing it. and not rely on AI.18

Developing legal instruments for data In recent years, we have seen noticeable activity
governance is also crucial. These instruments from governments and officials regarding policies
should provide oversight, establish limitations and regulations to address concerns about Big
on the power of technology companies, and Tech, with debates and actions still unfolding.
regulate how they utilize and share people’s data. In the US, for example, a group of lawmakers
proposed a legislative package to curb the market
Large technology companies such as Alphabet, power of dominant technology companies. This
Amazon, Apple, Meta, Microsoft, and Netflix follows investigations that accused the technology
possess vast amounts of confidential and giants of abusing their market power through
personal customer data. This data is acquired excessive fees, restrictive contracts, and data
from various sources, such as email and collection practices.19
business services, entertainment subscriptions
and viewing histories, shopping histories, and
location histories. It is imperative to introduce
guardrails on how such data can be utilized.
I think a global regulatory
Similarly, large financial institutions collectively
store extensive data encompassing billions of approach to AI is very
people. This data includes information about
individual’s purchasing habits. It is crucial that
desirable, because
clear limits are set on how this data can be used. I think we’re now at this
Social media platforms must also do a better point where these are
job at identifying and reducing the spread
of misinformation and disinformation. global challenges that
One strategy involves new governance
requirements, such as including public sector
require global norms
representatives on key boards to oversee AI and global standards
developments, data practices, and combatting
misinformation and disinformation. Satya Nadella20

29
Another example is the proposed US Senate
bill, The AMERICA Act (Advertising Middlemen
Endangering Rigorous Internet Competition
Accountability). This bill aims to prohibit large We have to ensure that AI
digital advertising companies from owning
both sides of an ad exchange, where they have
remains subject to human
ownership over demand and supply-side platforms control. Whether it’s a
— promoting a more competitive marketplace.21
government, the military
Additionally, President Joe Biden in late 2023
issued an executive order on AI. This order calls or any kind of organization,
for numerous federal agencies to set standards on
AI privacy, security and competition.22 Similarly, that is thinking about
in April 2023, the British Government unveiled its
Digital Markets, Competition and Consumer Bill,
using AI to automate, say,
a wide-ranging legislation aimed at improving critical infrastructure, we
consumer protection and fostering fair competition
in digital markets.23 need to ensure that we
In June 2023, European regulators filed antitrust have humans in control,
charges against Google. They argue that Google’s
dominant position in the digital advertising
that we can slow things
market can only be rectified by divesting parts down or turn things off
of its lucrative advertising business.24
Brad Smith25

Figure 13
AI-related bills passed into law globally
Number of AI-related bills passed into laws in 127 selected countries, 2016-23
40

35

30
28
25

20

15

10

0
2016 2017 2018 2019 2020 2021 2022 2023

Source: AI Index (2024) — processed by Our World in Data

30
Academia, Talent and R&D • Canada’s Global Innovation Clusters: These
Partnerships for Accelerated are pivotal in strengthening the nation’s
innovation landscape. The clusters include
Innovation and Diffusion
Digital Technology, Protein Industries Canada,
As AI reshapes the labor market, academic Next Generation Manufacturing Canada,
paradigms and curriculums must undergo Scale AI, and Canada’s Ocean Supercluster.
careful review. The goals of this review are They actively promote the adoption of
twofold: to cultivate talent with skills relevant domestically developed AI technologies across
to the AI-driven labor market and to reinforce various industries. In parallel, the Digital
Research Alliance of Canada provides AI
core competencies vital to society that may
researchers across the nation with dedicated
be neglected due to AI’s proliferation and its computing capacity
ability to perform certain tasks. Achieving these
objectives requires a coordinated effort involving
• Thakaa: In Saudi Arabia, the AI Center of
Advanced Studies, known as Thakaa, was
collaboration between government, academia,
established. This center serves as a hub for
and the private sector. advancing AI-related research and offers
valuable resources to researchers
To that end, efforts to prepare the current and
• University labs in Germany: Germany plans
future workforce for the age of AI are underway
to create 150 new university labs for AI
globally, with various initiatives focusing on research, expanding data centers and making
training and upskilling individuals to meet the the kind of complex public data sets from
demands of an AI-driven future. These include: which AI techniques can infer new insights
• The incorporation of data and AI programs more accessible28
into the formal education curriculums • Other examples: Large nations such as China
— such as Saudi Arabia’s MCIT Training and the United States also have specific
programs integrated into the Ministry of AI research programs driven by dedicated
Education curriculum agencies. In the United States, the Defense
• The creation of new executive degrees and Advanced Research Projects Agency (DARPA) is
certification programs — such as a graduate driving defense-related research. Additionally,
certificate at Stanford University, and courses various ‘moonshot’ programs have been
by UC Berkeley, and MIT xPro26 created in the areas of healthcare, energy, and
mobility, among others, beyond defense
• The launch of specific initiatives outside of
formal education curriculums — such as the A crucial aspect of these talent and R&D
UAE AI Summer Camp at which more than partnerships is the focus on accelerated general
22,000 participants received specialized
diffusion of AI within the economy. Collaborating
training on AI fundamentals27
with academia and the private sector, AI
• The launch of online and even free courses on technologies are initially applied to existing
fundamentals that can be accessed by anyone
processes and tasks to increase efficiency, reduce
across the globe — including offerings from
costs, and meet growing demand. Simultaneously,
Google, Microsoft, IBM, and LinkedIn
support for the development and deployment of
In parallel, coordinated scientific research complementary innovations is essential. These
and R&D programs, as well as R&D-enabling innovations will capitalize on AI and generative
institutions and organizations, should be AI technologies, such as natural language
established. These initiatives focus on accelerated interfaces that democratize access to technology,
innovation in areas that align with a nation’s to fundamentally reimagine how things are done
aspirations. Examples already underway include: and revolutionize current processes and operating
models. While these advancements are still
emerging, governments must proactively engage,
monitor, and steer related initiatives and the
ecosystem as a whole in an agile manner.

31
Infrastructure
Deploying AI applications requires significant Computing
infrastructure availability, including: If data serves as the fuel of AI, then hyperscale
cloud computing infrastructure, encompassing both
Data computing power and storage capacity, serves as its
Data and data quality have always been critical engine, alongside the critical AI algorithms. The rapid
enablers for improved business performance. deployment of supercomputers with exponentially
This is especially true for AI, where the adage of higher computing power is remarkable.
“garbage in, garbage out” rings truer than ever.
Data is the lifeblood of AI, the critical ingredient A notable example is Abu Dhabi’s G42 unveiling
to enable it to develop and continuously train its the world’s largest supercomputer for AI training
models. The lack of comprehensive and high- in partnership with Cerebras Systems, boasting a
quality data is a key driver of both AI bias and planned capacity of 36 exaFLOPS (36 x 10^18 FLOPS).
inaccuracies. According to a study published by The CEO of G42 Cloud highlighted the role of AI in
Vanson Bourne and Fivetran, underperforming enhancing productivity across various sectors as
AI programs/models built using low-quality or part of the UAE’s national AI plan, emphasizing the
inaccurate data cost companies up to 6% of annual potential for disruptive change.30
revenue or US$ 406 million on average.29
Recently, the US government imposed restrictions on
Both digitalization and the more recent proliferation the exports of state-of-the-art computing chips and
of Internet of Things (IoT) devices have led to an GPUs exceeding certain performance thresholds to
explosion of data availability across all aspects selected countries. Governments must evaluate such
of life. However, without proper enterprise data risks and proactively implement plans to mitigate the
governance and management frameworks that treat potential impact of geopolitical uncertainties.
data as a competitive business asset rather than an
IT issue, both the value of AI and the impact of AI Telecommunications
projects on business will be fully compromised. AI depends on robust telecommunications
infrastructure. Data and cloud computing require this
Beyond supporting the deployment of data infrastructure for seamless operation. Numerous AI
infrastructure and improving accessibility through applications require real-time data transfer between
open-data platforms, governments can establish sensors and the analytics engine, facilitated by
national standards and frameworks to promote cutting-edge telecommunications infrastructure
mature data management practices within such as fiber networks, 5G, and beyond.
organizations. Several countries, including Saudi
Arabia (with initiatives like the Saudi Open Data
Portal and Saudi National Data Management
Office), the UAE (Bayanat) and Singapore (data.gov.
sg), are already implementing such elements to
foster effective data management.

32
Foundation Models
Foundation models serve as the cornerstone custom-built AI foundation models. We observe
for AI systems, enabling them to undertake three prevalent approaches, as illustrated in
multiple tasks across different sectors of the Figure 14: private-led development with public
economy. Considering the extensive reach, support, public-led development with private
impact and associated risks of widespread AI support, and purely private-led initiatives.
adoption, particularly generative AI as previously Governments must evaluate their assets and
discussed, governments must consider the capabilities and determine the most suitable
development and operation of their own local approach for their circumstances.

Figure 14
Approaches to develop and operate foundational model

Private-led development Public-led development Private-led only

with public support with private support
Private entities lead the Government entities develop the Private entities work in silos
development of foundation foundation model, supported and do not benefit from public
models supported by by private companies providing support or provide their models
public entities resources and expertise to one another

ChatGPT-4 ALLaM Tongyi Qianwen

Mixtral 8x7B OpenGPT-X Krutrim LLM

Falcon 180B GPT-SW3 HyperCLOVA X

Note: These are example approaches — many countries have multiple approaches that sit across various categories, especially in leading tech
countries such as the US and China.
Source: Oliver Wyman analysis

Funding
As governments are employing a diverse range Europe’s total spend, and well ahead of China’s
of funding methods to drive AI research, US$ 7.76 billion. These figures underscore the
development, and innovation. These methods can importance of the private sector as an essential
be divided into budget allocations and specific driver of the AI revolution.33
funds. For instance, according to a report from
the Networking and Information Technology Moving on from budget allocations, examples of
Research and Development Program (NITRD), the specific funds include the Mohammed Bin Rashid
US government budget for AI research in 2024 is Innovation Fund in the UAE, which was allocated
US$ 3 billion.31 Germany plans to increase its public US$ 550 million to enhance the AI ecosystem
funding for AI research to €1.6 billion over the next and support local innovators.34 Another example
two years as it attempts to close a skills gap with is Denmark’s fund for digital welfare solutions
sector leaders China and the US.32 allocating tens of millions of dollars to, for
example, help municipalities and regions test new
These public sector amounts are however technologies for welfare services, or to establish
dwarfed by private AI spending in the US, which internet hotspots in public spaces.35
reached US$ 67.22 billion in 2023, six times

33
Funding is not only required for elements such as shift in AI development. This shift entails focusing
infrastructure and R&D for both AI development on building AI tools that are useful to human
and deployment, but also to steer and nurture workers rather than tools that will replace them.
the whole ecosystem in the desired direction. By adopting this approach, AI has the potential
As previously mentioned, there is a prevailing to contribute to shared prosperity rather than just
argument that emphasizes the need for a paradigm exacerbating the concentration of wealth.

How Should Governments Do It?

To effectively develop, oversee and deliver their understand the possibilities of real-world
AI strategies, governments need to first define applications, citizens who will adopt AI
their AI operating model at the national level. applications and who will ultimately bear much
This means a clear governance, mandate, and of the associated implications, international
assignment of ownership responsibilities. organizations working on AI standards and
transparency, and foreign governments aiming
At one end of the spectrum, exemplified by to realize AI benefits while mitigating potential
Singapore’s National AI Office, the operating negative effects.
model may adopt a centralized approach with
a national Center of Excellence or AI Authority For instance, G7 nations along with the EU have
responsible for developing, driving, and enforcing initiated the Global Partnership for AI (GPAI),
the AI strategy. Conversely, a distributed model an international initiative established to guide
mandates selected entities to develop and the responsible development and use of artificial
implement AI initiatives within their specific intelligence in a manner that respects human
domains, without centralized standards. rights and the shared values of its members.36

Hybrid or semi-centralized models, situated Meanwhile, the broader G20 has made
between these two extremes, blend elements AI a subject of discussions, focusing on
of both approaches. Typically, they involve AI principles and a human-centered
a central coordination unit for monitoring, approach to AI development. Key areas
standardization, and coordination. of stewardship outlined include inclusive
growth, sustainable development and
Once roles and responsibilities are assigned well-being; human-centered values and
to entities, each must proactively steer fairness; transparency; robustness, security
AI development and deployment efforts, and safety; and accountability. The UN also
encompassing aspirations, policies, regulations, created the “AI for Good Global Summit” to
talent, R&D, infrastructure, and funding. To serve as a platform for global AI dialogue.
achieve this, responsible entities must establish It is organized by the International
close collaborations, local, regional and Telecommunication Union (ITU) — the UN
international partnerships, and consultations specialized agency for information and
with all stakeholders in the AI ecosystem, both communication technology — in partnership
locally and internationally. with 40 UN sister agencies and co-convened
with the Government of Switzerland. It sets
Stakeholders include workers familiar with out to ensure that AI is employed to advance
the technology on the ground, policy-makers, health, climate, gender, inclusive prosperity,
researchers who understand the theoretical sustainable infrastructure, and other global
potential, tech-moguls who more closely development priorities.37

34
35
Conclusion

Governments worldwide view AI as a strategic technology, with

69 countries publishing national AI strategies according to the
OECD.38 Leaders acknowledge that global success hinges on AI
leadership. The potential advantages span industries and areas
such as economic health, job creation, social equity, public health,
food security, military capabilities, cybersecurity, personal privacy,
misinformation/disinformation, trust in governments, and the
climate crisis. AI applications have the potential to profoundly
impact all key areas of strategic national importance.

As the AI revolution unfolds, governments and policy-makers

must formulate a strategy to navigate its complexities and
strike a balance between benefits and drawbacks. This strategy
should encompass academic alignment, talent programs, public-
private collaboration, regulatory frameworks, and infrastructure
deployment with adequate funding.

To effectively oversee and implement this strategy,

governments need to define their national-level ‘AI operating
model’, which can take a centralized form, a distributed
model, or a hybrid of both approaches.

In conclusion, the global recognition of digital transformation and

the rapid advancements in AI underscore the need for governments
to formulate and implement comprehensive strategies to
harness its potential while mitigating risks. The future success
and competitiveness of nations will be shaped by their ability to
navigate the complexities of this transformative technology.

The race for AI will have winners and losers. It is incumbent upon
policy-makers to ensure that it becomes a major win and not an
existential loss.

36
37
About

Oliver Wyman

Oliver Wyman, a business of Marsh McLennan

(NYSE: MMC), is a management consulting
firm combining deep industry knowledge
with specialized expertise to help clients
optimize their business, improve operations
and accelerate performance. Marsh McLennan
is a global leader in risk, strategy and people,
advising clients in 130 countries across four
businesses: Marsh, Guy Carpenter, Mercer
and Oliver Wyman. With annual revenue of
$23 billion and more than 85,000 colleagues,
Marsh McLennan helps build the confidence
to thrive through the power of perspective.

Quotient – AI by Oliver Wyman

Quotient – AI by Oliver Wyman, leverages
decades of industry-specialized experience to
help our clients harness the value of AI at every
stage of their transformation journey. Quotient
brings together a global team of 1500+ data
scientists, engineers, strategists, and designers
who help our clients scale the power of AI in a
responsible manner, enabling organizations to
optimize operations, drive efficiencies, source
new value streams, and serve customers in
innovative and impactful ways.

For more information, visit oliverwyman.com

Follow Oliver Wyman on X @OliverWyman

38
39
References
1 Cerebras, Cerebras and G42 Complete 4 exaFLOP AI Supercomputer and Start the March Towards 8 exaFLOPs, 2023
2 Congressional Research Service, Generative Artificial Intelligence and Data Privacy: A Primer, 2023
3 CNBC, Elon Musk says AI will eventually create a situation where ‘no job is needed’, 2023
4 CNBC, Google CEO: AI is more important tan fire or electrivity, 2018
5 Thomas M. Siebel, Digital Transformation: Survive and Thrive in an Era of Mass Extinction, 2019
6 Oliver Wyman, The Transformation Of Aramco’s Commercial Operations, 2023
7 Can the Mere Mention of AI Move Stock Prices in 2023?
8 CNBC, Bill Gates explains why we shouldn’t be afraid of AI, 2023
9 OECD.AI Policy Observatory, National AI policies & strategies, 2024
10 AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity
11 CNBC, Goldman Sachs says generative A.I. could impact 300 million jobs, 2023
12 Daron Acemoglu, The Simple Macroeconomics of AI, NBRE, 2024
13 CNBC, New York Times sues Microsoft, ChatGPT maker OpenAI over copyright infringement, 2023
14 NIST, NIST Study Evaluates Effects of Race, Age, Sex on Face Recognition Software, 2019
15 NBC News, Black plaintiffs file lawsuits for wrongful arrests or jailing due to facial recognition technology, 2023
16 Stanford University. “Machine learning (ML) platform emissions in CO2 equivelant in 2022, in tonnes.” April 15, 2023. Statista
17 CNBC, Warren Buffet says AI scammming will be the next big “growth industry”, 2024
18 NBC, ChatGPT banned from New York City public schools’ devices and networks, 2023
19 CNBC, Lawmakers unveil major bipartisan antitrust reforms that could reshape Amazon, Apple, Facebook and Google, 2021
20 CNBC, Microsoft CEO Satya Nadella says global consensus on AI is emerging, 2024
21 Congress.gov, America Act, 2023
22 The White House, President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence, 2023
23 UK Parliament, Digital Markets, Competition and Consumers Act 2024, 2024
24 European Commission, Antitrust: Commission sends Statement of Objections to
Google over abusive practices in online advertising Technology, 2023
25 CNBC, AI needs human control to avoid bieng weaponized, says Microsoft’s president, 2023
26 Study Portals, Bachelor’s degrees in Artificial Intelligence, 2024
27 UAE Minister of State for AI, UAE AI Camp, 2024
28 BMBF, AI Action Plan, 2023
29 Fivetran, New AI survey: Poor data quality leads to $406 million in losses, 2024
30 The National News, Abu Dhabi’s unveils world’s largest supercomputer for AI training, 2023
31 NITRD, AI R&D Investments, 2024
32 BMBF, AI Action Plan, 2023
33 Stanford University, Private investment in AI worldwide in 2023 by country, Statista, 2024
34 UAE Minister of State for AI, AI National Strategy, 2023
35 Danish Agency for Digital Government, The Danish Strategy for AI, 2019
36 OECD AI Policy Observatory, Global Partnership on AI, 2021
37 UN, AI For Good Summit, 2023
38 OECD, AI Policy Observatory, 2024

40
Authors
Jad Haddad
Partner and Global Head of Quotient - AI by Oliver Wyman
[email protected]

Abhishek Sharma
Partner, Oliver Wyman
[email protected]

Amir Salam
Principal, Oliver Wyman
[email protected]

41