ARTIFICIAL INTELLIGENCE IN GOVERNMENT
KNOWLEDGE BASED ECONOMY RESEARCH STUDY PREMK BRIN (22 AUG 2024)
Introduction
Artificial Intelligence (AI) is rapidly transforming
various sectors, and the government is no exception.
AI offers immense potential to improve public services,
optimize
operations,
and
enhance
citizen
engagement. Its ability to analyze massive datasets,
automate tasks, and provide personalized solutions
makes it a powerful tool for modern governments.
AI applications have already been adopted by various
public agencies and organizations that implement
government policies and may contribute to their
development to make government services more effective
and responsive.
Benefits of AI in Government
1
3
Improved Efficiency
2
Enhanced Decision-Making
AI can automate repetitive tasks, freeing up government
AI algorithms can analyze vast amounts of data to identify
employees to focus on more complex and strategic initiatives.
trends and patterns, providing insights that can inform better
This leads to faster service delivery, reduced costs, and
decision-making. This leads to more informed policies, efficient
improved overall efficiency.
resource allocation, and improved outcomes for citizens.
Personalized Services
4
Increased Transparency
AI can personalize public services based on individual needs
AI can be used to streamline and automate processes, making
and preferences. This results in more relevant and responsive
government operations more transparent and accountable. This
government interactions, leading to increased citizen
builds trust between the government and the public.
satisfaction and engagement.
Challenges of AI Implementation
Data Privacy and Security
Algorithmic Bias
Job Displacement
Government data is often sensitive,
AI algorithms can reflect biases
AI automation could potentially
so ensuring data privacy and security
present in the data they are trained
displace some government jobs. It's
is paramount when implementing AI.
on. This can lead to discriminatory
essential to consider retraining and
Robust security measures are
outcomes, especially in areas like
upskilling programs to prepare
needed to protect sensitive
justice, healthcare, and employment.
employees for the evolving job
information from unauthorized access
Addressing algorithmic bias is crucial
market and ensure a smooth
or breaches.
to ensure fairness and equity.
transition.
Ethical Considerations
Transparency
Fairness
AI systems should be transparent in
AI systems should be designed to be
their decision-making processes.
fair and equitable, avoiding bias and
Citizens should understand how AI
discrimination. This ensures that all
algorithms arrive at their conclusions,
citizens receive equal opportunities
promoting accountability and trust.
and benefits from AI-powered services.
Accountability
Privacy
There should be clear mechanisms for
AI systems should respect citizens'
holding those responsible for AI
privacy and handle personal data
development and implementation
responsibly. This involves
accountable for any unintended
implementing robust data protection
consequences or ethical violations.
measures and ensuring transparency
about data collection and usage.
Developing an AI Strategy
Needs Assessment
1
Identify specific challenges and opportunities where AI can
create value for the government. This involves
understanding current processes, data availability, and
2
citizen needs.
Pilot Projects
Start with pilot projects in specific areas to test the
feasibility and effectiveness of AI solutions. This allows for
Infrastructure Development
experimentation, learning, and refinement before large-
3
scale implementation.
4
Stakeholder Engagement
Develop the necessary infrastructure, including data
storage, processing power, and security measures, to
support AI applications. This involves ensuring data quality,
scalability, and security.
Engage with relevant stakeholders, including citizens,
government agencies, and technology experts, to build
Continuous Evaluation
Continuously monitor the effectiveness and impact of AI
applications, adapt strategies based on emerging
technologies and changing needs, and ensure ongoing
ethical review and accountability.
5
consensus, address concerns, and ensure responsible AI
implementation.
THE AI LIFECYCLE
(Source: Trail, 2024)
Integrating AI with Existing Systems
Data Integration
Ensure seamless data flow between AI systems
and existing government databases, maintaining
data integrity and security.
API Development
Develop Application Programming Interfaces
(APIs) to allow AI systems to interact with existing
government applications, enabling efficient data
exchange and communication.
System Interoperability
Ensure that AI systems can interact with different
government platforms and services, promoting
seamless data sharing and collaboration.
User Interface Design
Design user-friendly interfaces for AI-powered
applications, making them accessible to
government employees and citizens alike.
PROPOSED DIMENSIONS TO ANALYSE AND CLASSIFY AI SYSTEMS
IN GOVERNMENT
Dimension
Definition
Operational fitness
The degree to which the composition and functions of a government system
incorporating an AI application (or set of applications) aligns with (1) codified
standards of organisation, system construction and functioning required to operate in
a particular environment, and the extent to which it can be expected to (2)
outperform other agents at a specific cognitive task (or set of tasks).
Epistemic alignment
The degree to which (1) data and information about the composition of a government
system incorporating an AI application (or set of applications) align with standards of
knowledge sharing, and (2) its behaviour in every circumstance is known by all affected
parties.
Normative divergence
The degree to which the observed behaviour of a government system incorporating
an AI application (or set of applications) in its current environment, including the
consequences of the system’s behaviour, diverges from (1) formal institutional
standards and (2) affected parties’ perceptions of acceptable behaviour.
(Source: V.J. Straub et al., 2023)
CHARACTERISTICS OF PROPOSED DIMENSIONS TO CLASSIFY AIBASED SYSTEMS IN GOVERNMENT (1)
Dimension
Operational
fitness
Dimension
levels
Basic
Intermediate
Advanced
Description
Example measurement standards
The AI-based system meets critical benchmarks in core •
operational fitness standards, especially in terms of
effectiveness. Application components satisfy core
•
Objectives
•
The AI-based system exceeds critical benchmarks in
core operational fitness standards, especially in terms
of effectiveness and safety. Application components
•
satisfy core objectives
The AI-based system meets or exceeds benchmarks in
all operational fitness standards, including in terms of
effectiveness, safety, reliability, usability, and adaptability.
Application components satisfy all Objectives
(Source: V.J. Straub et al., 2023)
•
ISO/IEC DTS 4213.2 Assessment Of
Machine Learning Classification Performance
Singapore AI Governance Testing
Framework Toolkit
ISO/IEC TR 24029–1:2021 Artificial
Intelligence (AI) — Assessment of the
robustness of neural networks
ISO/IEC TR 24027:2021 Information
technology — Artificial intelligence (AI) —
Bias in AI systems and AI aided decision
making
Model cards for model reporting in Mitchell
et al., 2019
CHARACTERISTICS OF PROPOSED DIMENSIONS TO CLASSIFY AIBASED SYSTEMS IN GOVERNMENT (2)
Dimension
Epistemic
alignment
Dimension
levels
Nominal
Dispersed
Description
Example measurement standards
Data and information about the AI-based system meets •
critical benchmarks in core epistemic alignment standards,
especially in terms of transparency. Knowledge about the •
system is incomplete
Data and information about the AI-based system exceeds •
UK Algorithmic Transparency
Standard
UK Metadata Standards for Sharing
and Publishing Data
IEEE Standard for XAI – eXplainable
Artificial Intelligence - for Achieving
Clarity and Interoperability of AI
Systems Design
IEEE 7001– Standard for
Transparency of Autonomous Systems
Datasheets in (Mitchell et al., 2019)
critical benchmarks in core epistemic alignment standards,
especially in terms of transparency and explainability. •
Knowledge about the system is distributed unequally among
•
affected Parties
Uniform
Data and information about the AI-based system meets or •
exceeds benchmarks in all epistemic alignment standards,
including in terms of transparency, explainability,
reproducibility, interpretability, and openness. All relevant
knowledge about the system is distributed appropriately
among affected Parties
(Source: V.J. Straub et al., 2023)
CHARACTERISTICS OF PROPOSED DIMENSIONS TO CLASSIFY AIBASED SYSTEMS IN GOVERNMENT (3)
Dimension
Normative
divergence
Dimension
levels
Low
Moderate
High
Description
Example measurement standards
The behaviour of the AI-based system consistently converges •
with codified standards and affected parties’ perceptions of
acceptable behaviour
•
•
The behaviour of the AI-based system consistently converges •
•
with codified standards of acceptable behaviour, but
there is variance in affected parties’ perceptions
•
Algorithm Charter for Aotearoa New
Zealand
UK Public Attitudes to Data and AI
Tracker Survey
EU AI Alliance
Canadian Algorithmic Impact
Assessment Tool
ISO/IEC 38507:2022 Information
technology — Governance of IT —
Governance implications of the use of
artificial intelligence by organisations
ForHumanity Independent Audit of
AISystems (IAAIS)
The behaviour of the AI-based system consistently diverges
from codified standards of acceptable behaviour. Operational
fitness or epistemic alignment may need to be re-evaluated. If •
normative divergence remains universal, other changes may
have to be made
(Source: V.J. Straub et al., 2023)
THANK YOU
Research Center for Macroeconomics and
Finance OR TKPEKM - BRIN
Gedung Widya Graha Lantai 9, Jl. Gatot
Subroto 10, Jakarta 12170, Indonesia