The document discusses opportunities for using artificial intelligence (AI) in the public sectors of Western Balkan countries. It provides examples of AI use cases from European public administrations, such as using satellite data to detect mowed grasslands in Estonia. It also discusses factors that influence successful AI adoption, such as developing internal expertise, focusing on use cases aligned with organizational needs, and considering long-term adoption throughout the project lifecycle. The document emphasizes that AI adoption requires organizational transformation, not just implementing technology solutions.

1. Opportunities of AI for Western-
Balkan public sector
Overview and Potential of AI for Government
Colin van Noordt
Programme Director AI - Digital Nation
PhD Researcher - TalTech
colin.vannoordt@digitalnation.eu
06-06-2023

2. Who am I?
• Programme Director for AI – Digital Nation
• PhD Researcher at Ragnar Nurkse Department of Innovation and
Governance at TalTech
• Former External Expert to the European Commission’s AI Watch research
team on AI in the public sector
• Project and research assistant IntouchAI.EU Project
• Co-Editor of the Research Handbook on Artificial Intelligence for Public
Management published by Edward Elgar Publishing
• Upcoming Researcher Digitalisation at the Dutch Court of Audit
(September 2023)

4. The roles of government in AI
As Facilitator As Regulator As User

5. We still know very little on AI in government
1 2 1 1 2
22 23
56
47
77
1999 2002 2005 2016 2017 2018 2019 2020 2021 2022
Research on Artificial Intelligence in
Digital Government Research
5 million AI publications in 2021 Almost none in digital government (0.00464%)

6. But what is AI? Futuristic robots…
“the hardware capacity for human-equivalent artificial
intelligence will likely exist before the end of the first
quarter of the next century, and may be reached as early
as 2004.” – Bostrom 1998

7. Or current applications?
• Da Vinci Surgical System
Developments in self-driving cars
Precision farming
Large-scale language models

8. Technical perspectives

9. Supervised learning

10. Unsupervised learning

11. Reinforcement learning

12. The humans behind AI

13. AI for policymaking
SATIKAS to detect mowed grass of
farmers, Estonia
The Dublin Beat
analyses citizen
tweets
Object Detection, Amsterdam
CitizenLab to analyse
citizen input
AI to improve various stages of policy
making
- Detecting social issues more quickly
- Estimate potential effects of policy
options
- Improve and fasten decision making
- Monitor ongoing implementation of
policy
- Evaluate existing policy
- Include citizens in policymaking

14. AI for public service delivery
BüroKratt AI, Estonian
Government
Misty II to assist the elderly in
Barcelona
JobBereik to assist in reskilling, VDAB,
Belgium
AI could be used to deliver public services to
businesses and citizens
- Enhance information delivery about government
services
- Improve public services to citizens and
businesses, through personalization
- Automate redundant processes and reducing on-
site meetings
- Develop completely new services through AI
- Reduce corruption and improve trust in public
service delivery
- Empower civil servants through decision support
tools

15. AI for internal management purposes
Tengai interviewing job
applicants, Sweden
AI to detect anomalies
in X-Road, Estonia
VeriPol to detect false police
reports, Spain
AI to improve internal management
operations
- Improve recruitment services
- More efficient allocation of human
resources
- Improved financial management
- Strengthen cybersecurity
- Predictive maintenance
- Modernize public procurement
processes
- Improve detection of fraud

16. The gap between technological progress and practice

17. AI in government = digital transformation
"The technology community in government is often expected to
drive transformation on behalf of business leaders. However,
most digital change decisions are made by the business leaders –
such as permanent secretaries, chief executives, chief operating
officers and directors general.(…) The success of these decisions is
dependent on these leaders having the digital fluency to make the
best choices and fully understand the consequences of their
decisions for digital transformation.“ – UK National Audit Office,
March 2023

18. Success factors

19. Public sector context

20. Avoiding technosolutionism

21. Importance of internal expertise and
competences

22. Ethical and legal guidelines

23. Do not forget the end-users
“We don’t only need to be good at predicting but
also at explaining to the consultants. If you do not
have the domain knowledge, you do not know the
business process at well. Then you can create a
perfectly accurate model at prediction that
provides completely useless information.”

24. Minding the gap between a pilot and
implementation
Scope of the project
Funding and expertise
Responsibility and risk
Development and maintenance

25. Looking for the best practices
Think big, but start small
Partners, not clients of the private sector
Collaboration from the start
AI project management practices
Demystify expectations of AI – both good and bad
Ethical and legal risks tackled from the start
Take into account the bigger picture of implementation

26. We are still at the beginning of the
AI-journey in government. Let’s
explore it together.
Digital Nation
digitalnation.eu
info@digitalnation.eu
LinkedIn
57 Telliskivi, Tallinn
Harju maakond, 10412
Estonia
Colin van Noordt
Programme Director AI - Digital Nation
PhD Researcher - TalTech
colin.vannoordt@digitalnation.eu

27. Opportunities of AI for Western-
Balkan public sector
Use cases from European public administrations
Colin van Noordt
Programme Director AI - Digital Nation
PhD Researcher - TalTech
colin.vannoordt@digitalnation.eu
06-06-2023

28. AI Watch research

29. AI Strategies in EU
AI strategies in the European Union (by Feb 2022, source JRC)

30. Key strategic actions to boost AI in government
Improving data access and management
Ethical AI design principles
Facilitating partnerships with the private sector
Improving internal competences
Improving knowledge and awareness of AI
Introduction and review of legislation
New teams or institutions for AI in the government

31. AI strategy initiatives
Improving data access
and management
Czechia
Developing a binding public
administration data
availability plan for AI use,
including data standards.
Ethical AI guidelines
Austria
For this purpose, framework
conditions
are created that ensure compliance
with human rights and security of the
use of AI during the entire technical
life cycle

32. AI strategy initiatives
Partnerships with private
sector
Ireland
Instruments such as dialogues, hackathons
and pre-commercial procurement of
innovative solutions will enable suppliers to
respond better to public procurement
requests and also assist public authorities to
understand the market better and formulate
targeted procurements.
Improving internal
expertise
France
Public authorities must therefore
equip themselves with the skills required
to better understand, identify and fight
against forms of algorithmic
discrimination, particularly when it
affects access to basic services

33. AI strategy initiatives
Awareness campaigns
Estonia
Organizing the spread of
knowledge and exchange of
experience - to introduce the
possibilities and examples of
AI in different networks and
formats
New legislation for AI
The Netherlands
Companies and governments
have a (legal) responsibility to
provide adequate insight into
the (procedure surrounding) AI
applications they use.

34. What AI strategies are missing
Limited focus on AI in government
Excessive focus on data
Internal capacity overlooked
Funding often missing
Organisational AI strategies

35. High variety of AI adoption rates
Enterprises using AI technologies by size class, EU, 2021 (Eurostat)
Digitalisation in Europe 2021-2022, European Investment Bank (2022)
Digital Innovation in Governance and Public Service Provision (DIGISER), 2022
Gartner CIO Survey 2021

36. Core findings
9%
10%
27%
54%
0% 20% 40% 60%
Multinational
Regional
Local
National
Public services
and
engagement
35%
Enforcement
25%
Analysis,
monitoring and
regulatory
research
22%
Internal
management
16%
Adjudication
2%
PURPOSE OF AI
3%
4%
25%
30%
38%
Planned
Not in use anymore
In development
Pilot
Implemented

37. Are government organisations ready for AI?

38. Key adoption challenges
The use of new innovations, such as ICT, is not
straightforward in government
- Technological challenges
- Laws and regulation
- Ethics
- Society
- Other factors such as funding, perceived value…
- Cultural
Some governments still need to invest massively in
‘basic’ digital government

39. Learning from examples
Agricultural Registers and
Information Board (ARIB)
Estonia
SATIKAS
Detection of the mowing of grasslands with COPERNICUS
Description
The AI system combines machine learning methods to
analyse satellite data together with other data sources. Used
to optimize inspection capacity of ARIB and enforcement of
subsidy requirements.
Lessons learned
Collaboration and sharing of resources crucial to ensure
adoption and implementation. There can be a significant
amount of time between research and actual implementation.
Kind en Gezin agency
Flemish State, Belgium
Description
Aim of the system is to optimize inspection capacity
and enhance inspection practices using a variety of
data sources. Enables more targeted interventions.
Lesson learned
Civil servants need to see the system as an
empowerment, rather than a replacement of their
knowledge.
Predictive system day-care services
Predictive model to detect day-care services in need of
further inspections

40. Learning from examples
Danish Business Authority
Denmark
Description:
The Danish Business Authority has been using a variety
of AI models to detect fraud and tackle financial crimes,
coming together in the Intelligent Control Platform.
Unique insight:
XRAI Methodology has been used to ensure that a
precise description of the needs and expectations of the
business is available. Strong data governance practices
ensure quality and trust in system.
Intelligent Control Platform
Combating business fraud
Greek Government
Greece
Description:
During the COVID-19 Crisis, the Greek government has
been using an AI system in all border control points
which helps the selection of which travellers to test upon
arrival at the border.
Unique insight:
Eva was not developed in a mathematically optimal way
but was designed to be practical, effective, transparent
and explainable. Black-box algorithms were avoided to
avoid opacity.
Eva
Targeted COVID-19 Border Checking

41. Learning from examples
Amsterdam
the Netherlands
Description:
The city of Amsterdam used the Object Detection Kit to
detect identify thrash in the cities, such as graffiti, trash,
broken traffic lights and other day-to-day issues which
plague livelihood of the city.
Lessons learned:
The Amsterdam Object Detection Kit was not put in use
following the pilot as the focus of the initiative was the
develop rather than implement it. Funding was not
available for implementation.
Object Detection Kit
Detecting garbage and other objects in the city
Estonian Unemployment Insurance
Fund
Estonia
Description:
An AI system to assist new staff with predicting chances
of citizens to get a new job after unemployment.
Lessons learned:
There is a close collaborative culture where new
innovations are openly shared and discussed. Staff
members to understand what AI is and come up with
suggestions to improve their work.
OTT
Speech recognition tool in the Estonian Parliament

42. Learning from examples
Register for Enterprises
Latvia
UNA
Chatbot supporting citizens
Description
UNA is a Chatbot able to answer frequently asked
questions about the registration of their businesses as well
as the liquidation, merchants, companies and organizations
and their application processes.
Lessons learned
It is said that 44% of the questions asked can easily taken
care of by the Chatbot. Other non-standard issues are still
handled by the support staff, who now have more time for
complex requests
Spanish National Police
Spain
Description
The AI helps tackle the police in wasting resources on
false police reports. Has been regarded as useful by the
staff, leaving them with more time to focus on other tasks
Lessons learned
Successful pilot projects need to be scaled up to ensure
adoption throughout multiple administrations
VeriPol
AI system to detect false police reports

43. Crucial take-aways
• Factors contributing to the adoption of AI are not the same as those
improving performance/public value
• Grow the government with the ambitions for AI
Adoption =/= improvement
• It is simpler to do data analysis than it is to change the organisation
• Transformation comes from public sector leaders, not tech vendors
Use of ML models =/= Use of AI
system in organisational context
• Initial adoption may be successful yet AI adoptions may end nonetheless
• Think strategically about the use of AI – no gimmick technology
Consider lifecycle and long-term
adoption as well
• Those who are likely to benefit the most from AI suffer from the most barriers
• Different AI systems may be relatively easier to adopt than others
Maturity / readiness of public
organisations may differ
substantially

44. We are still at the beginning of the
AI-journey in government. Let’s
explore it together.
Digital Nation
digitalnation.eu
info@digitalnation.eu
LinkedIn
57 Telliskivi, Tallinn
Harju maakond, 10412
Estonia
Colin van Noordt
Programme Director AI - Digital Nation
PhD Researcher - TalTech
colin.vannoordt@digitalnation.eu