Content uploaded by Tzuhao Chen
Author content
All content in this area was uploaded by Tzuhao Chen on Mar 01, 2025
Content may be subject to copyright.
PUBLIC PERFORMANCE & MANAGEMENT REVIEW
Holding AI-Based Systems Accountable in the Public
Sector: A Systematic Review
Qianli Yuana and Tzuhao Chen
b
aFudan University; bUniversity at Albany – SUNY
ABSTRACT
Due to the significant advancements in technology and com-
puting power over the past few decades, artificial intelli-
gence-based systems have emerged as desirable tools for
improving public services. Scholars have dedicated consider-
able effort to understanding this growing trend. A key focus of
this research has been accountability, which has become a
prominent concern in the scholarly discourse. Studies reveal
that the integration of AI into government processes intro-
duces unique challenges to maintaining governmental account-
ability. In response, scholars have proposed various policy and
management strategies to tackle these issues. Despite exten-
sive research, there remains a divergence in how accountability
is conceptualized in the context of using AI-based systems in
government, indicating a gap in the systematic understanding
regarding how to hold the use of such systems accountable. To
bridge this gap, we conducted a systematic review of the exist-
ing research on AI-based system accountability in the public
sector. Our findings based on thirty-six articles highlight the
existence of different forums and actors, and the mechanisms
recommended to ensure accountability. We also identified chal-
lenges and enablers that could affect the efficacy of these
mechanisms. Lastly, our analysis uncovers several areas requir-
ing further exploration, offering directions for future research in
this field.
Introduction
The substantial advancements in technology and computing power over
recent decades have positioned artificial intelligence (AI) as a pivotal tool
for addressing citizens’ information needs and enhancing public service
delivery. Scholars in public administration and other related fields have
made significant efforts to comprehend this escalating trend. Research, in
general, indicates that AI can be applied across various policy areas and
tasks (Chung et!al., 2022; de Sousa et!al., 2019), offering numerous ben-
efits for public values (Henman, 2020; Pencheva et!al., 2020; Wirtz et!al.,
https://doi.org/10.1080/15309576.2025.2469784
© 2025 Taylor & Francis Group, LLC
CONTACT Tzuhao Chen tchen9@albany.edu Rockefeller College of Public Aairs and Policy, University at
Albany – SUNY, 135 Western Avenue, Albany, NY 12203, USA
KEYWORDS
Articial intelligence;
accountability;
algorithmic
accountability; AI
governance; transparency
2 YUAN AND CHEN
2021). However, the integration of AI-based systems also introduces various
ethical, political, social, and legal challenges and concerns for public orga-
nizations and the broader society (Wirtz et!al., 2021; Zuiderwijk et!al.,
2021). Therefore, it’s crucial to ensure that AI-based systems in the public
sector are used in a way that mitigates these pitfalls and maximizes their
intended benefits.
In particular, accountability has emerged as a critical issue that garners
scholarly attention. Numerous studies indicate that the introduction of
AI-based systems has brought about unique challenges to government
accountability (Bracci, 2023; Busuioc, 2021). This is primarily due to the
use of AI-based systems in decision-making processes, which may alter
the roles and responsibilities of public employees, leading to ambiguity
regarding who or what should be held accountable (Bannister & Connolly,
2020; Donovan et!al., 2018). Furthermore, the inherent characteristics of
AI-based systems, such as their opaqueness and complexity, pose additional
challenges in ensuring accountability (Bracci, 2023; Busuioc, 2021).
Considering these challenges, scholars across various fields have put forth
several policy and management recommendations to address the issue. In
short, these recommendations seek to ensure that the use of AI-based
systems by public organizations is transparent, explainable, and accountable
(Ada Lovelace Institute et!al., 2021; Bracci, 2023; Busuioc, 2021;
Grimmelikhuijsen, 2023).
Although there has been extensive research on the topic, the literature
reveals diverse definitions and conceptualizations of accountability in the
context of AI-based systems in the public sector, highlighting a lack of
systematic understanding. This study aims to clarify the current state of
research by addressing three critical questions: (1) Who are the primary
stakeholders involved in the accountability process of AI-based systems? (2)
What mechanisms can enhance accountability in the use of AI-based systems?
and (3) What factors either enable or challenge these mechanisms? Our
overarching goal is to develop a comprehensive research agenda that
advances beyond the existing scope of accountability in AI-based systems
within governmental contexts. To achieve this, we performed a systematic
literature review, synthesizing and consolidating knowledge in this field
to create a more coherent understanding about the literature. This approach
not only allows us to integrate findings from previous studies but also
fuels scholarly debate and lays the groundwork for future research
directions.
This article is structured as follows: Initially, we present a brief over-
view of the concepts of accountability and artificial intelligence.
Subsequently, in the research design section, we outline the methodology
used for conducting the systematic review. The following analysis section
is divided into two parts: the first part explores the basic information
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 3
for the included literature, while the second part delves into the mech-
anisms, enablers, and challenges related to the accountability of AI-based
systems in the public sector. We conclude the article by discussing the
implications derived from our review findings and presenting a
research agenda.
Background: Government accountability and articial intelligence
The research on government accountability in modern contexts finds its
origins in the debate over administrative responsibility in democratic
governments during the 1940s (Finer, 1941; Friedrich, 1940; Plant, 2011;
Ya n g & D u b n i ck , 2016). In this debate, Carl J. Friedrich emphasized the
importance of internal checks and professionalism, while Herman Finer
advocated for external control. Since then, interest in and study of account-
ability have surged, driven by the increasingly complex and intricate chal-
lenges faced by governments worldwide, which have significantly reshaped
governance structures and the dynamics among public sector, private sector,
and society (Dubnick, 2014). Notably, government accountability is rec-
ognized as a multi-dimensional concept, leading to varied interpretations
across different academic disciplines (Bovens, 2014). Among the various
definitions, the one provided by Bovens (2007) offers a comprehensive
view of government accountability: “A relationship between an actor and
a forum, in which the actor has an obligation to explain and to justify his
or her conduct, the forum can pose questions and pass judgement, and the
actor may face consequences (p. 447).”
The definition above elucidates the essential elements of government
accountability. Building upon this definition, it is advised to address
several pivotal questions in any study of accountability: Who is being
held accountable and to whom? for what are they accountable? by what
standards are they judged? and why is this accountability necessary?
(Bovens, 2014). These questions, when answered, can reveal the multi-
faceted nature of accountability. Current research has identified diverse
forms of accountability (see Aleksovska et!al., 2019). For instance, Romzek
& Dubnick (1987) categorize accountability based on the source and
degree of control, identifying political, legal, bureaucratic, and profes-
sional accountability as key types. Besides, distinctions are made between
formal and informal types of accountability particularly in the context
of network governance (Romzek et!al., 2012). Boven’s framework empha-
sizes external control as a critical aspect of accountability, distinguishing
it from responsibility, which is more oriented toward internal control.
However, it is important to note that some scholars consider responsi-
bility and internal checks as another form of accountability, such as
professional accountability.
4 YUAN AND CHEN
Artificial intelligence (AI) can be understood as a machine-based system
that exhibits intelligent behavior by analyzing its environment and taking
actions, with some degree of autonomy, to achieve human-defined objectives
(European Commission, 2018; OECD, 2019; Russell, 2019). Such systems
can make predictions, recommendations, or decisions that influence real or
virtual environments. The concept of AI is not new, but it is the recent
advancements in machine learning and big data that have brought AI into
the spotlight (Jordan, 2019; Pencheva et!al., 2020). These advancements
enable AI-based systems to effectively learn patterns, from both structured
and unstructured data, and enhance their ability for self-learning. The
enhanced capabilities of AI have made it more powerful and versatile,
enabling its application across a broader spectrum of fields and industries.
The growing interest and actual deployment of AI in the public sector
worldwide have escalated significantly in recent years. AI can be associated
with various government functions, such as general public service, eco-
nomic affairs, and environmental protection, among others (de Sousa et!al.,
2019). In practical applications, it is found that public organizations fre-
quently utilize AI-based systems such as chatbots, predictive analytics,
computer vision, identity recognition, and expert systems for decision-mak-
ing support (van Noordt & Misuraca, 2022). The existing literature in
public administration has identified several themes related to the use of
AI in government, including challenges and strategies in AI adoption and
implementation (Chen et!al., 2024; Madan & Ashok, 2023; Nam & Bell,
2024), the relationship between AI and public values (Chen et!al., 2023;
Criado & Gil-Garcia, 2019; Wirtz et!al., 2019, 2020), and the dynamics
between AI-based systems and human public employees and the implica-
tions (Bullock, 2019; Bullock et!al., 2020; 2022; Dunleavy & Margetts,
2023; Young et!al., 2019).
The impact of AI-based systems on government accountability is a
pivotal area of scholarly focus. At its core, the accountability of AI-based
systems pertains to the relationship between account givers and takers,
where the account givers are required to explain and justify their decisions
informed by AI-based systems and face consequences (Bannister &
Connolly, 2020; Bracci, 2023; Busuioc, 2021; Donovan et!al., 2018; Williams
et!al., 2022). It is suggested that the deployment of AI-based systems can
lead to adverse effects on populations, including issues related to trust,
privacy, ethics, biases, and inequality. Consequently, ensuring that the
relevant parties are held accountable for their use of AI-based systems is
of paramount importance (Bracci, 2023; Chen et!al., 2023; Henman, 2020).
Broadly speaking, the major assumption behind this discussion is that AI’s
actions can potentially deviate from human objectives. Therefore, it is
necessary to ensure that AI remains under human control, so that it will
serve human preferences (Russell, 2019).
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 5
To e ns u re a c co u nt ab il i ty i n t he u s e o f A I- ba s e d s ys t em s , t hr ee i n te r-
connected dimensions are emphasized in the literature, including: (1)
Information, which involves the transparency of conduct, performance,
and the resources utilized. (Bracci, 2023; Busuioc, 2021; Fink, 2018; Giest
& Grimmelikhuijsen, 2020; Robinson, 2020; Selten & Meijer, 2021); (2)
Explanation (or justification/answerability), which refers to the process of
explaining and justifying how outputs are achieved (Bracci, 2023; Busuioc,
2021; Grimmelikhuijsen, 2023; Robinson, 2020; Selten & Meijer, 2021);
and (3) Consequences, which means imposing sanctions and requiring
actors to take remedial actions to address failures and provide redress to
those adversely affected (Ada Lovelace Institute et!al., 2021; Bannister &
Connolly, 2020; Bracci, 2023; Busuioc, 2021; Williams et!al., 2022).
This systematic review aims to enhance our understanding of govern-
ment accountability, particularly in the context of using AI-based systems
in the public sector. First, it systematizes existing knowledge about the
primary stakeholders—both account givers and takers—involved in the
accountability process. Second, it identifies and evaluates various mecha-
nisms, such as strategies and policies, that have been proposed or imple-
mented to bolster accountability in AI applications. Third, the review
investigates factors that either facilitate or hinder these mechanisms’ effec-
tiveness. By analyzing these mechanisms, along with their enablers and
challenges, the review offers a critical perspective on the effectiveness of
the mechanisms and the areas needing further development. Finally, we
intend to identify gaps in current research regarding the accountability of
AI-based systems and suggest future research directions for deeper under-
standing of the dynamics. Overall, our review findings could serve as a
valuable resource for academics, policymakers, and practitioners to com-
prehensively and systematically understand the accountability of AI-based
systems in the government context.
Research methodology
The systematic literature review approach was utilized to answer the
research questions. The systematic review is a method where researchers
adopt a few principles to ensure structuredness, comprehensiveness, and
transparency in selecting studies, assessing contributions, analyzing and
synthesizing data, and finally reporting the evidence (Denyer & Tranfield,
2009; Hiebl, 2023). Hence, a systematic literature review “allows reasonably
clear conclusions to be reached about what is and is not known” (Denyer
& Tranfield, 2009, p. 671). This approach will help clarify our current
understanding of the accountability of AI-based systems in the public
sector and identify areas that require further scholarly investigation. To
ensure reproducibility and transparency of the literature review, this
6 YUAN AND CHEN
research follows the Preferred Reporting Items for Systematic Reviews and
Meta-Analyses (PRISMA) protocol to implement the literature review in
the following steps: identification, screening, eligibility, and inclusion
(Liberati et!al., 2009), as indicated in Figure 1.
In the identification phase, we sought to search for all the studies that
have to do with the accountability of AI-based systems in the public
sector; hence, we employed the following search words with Boolean
operators to identify the literature: ("Artificial intelligence" OR "AI" OR
“artificial general intelligence” OR “AGI” OR "machine learning" OR "deep
learning" OR "reinforcement learning" OR "supervised learning" OR "unsu-
pervised learning" OR "neural networks" OR "natural language processing"
OR "computer vision" OR "image recognition" OR "facial recognition"
OR "face recognition" OR "speech recognition" OR "intelligence systems"
OR "virtual assistant" OR "autonomous vehicle" OR "predictive analytics"
Figure 1. PRISMA ow chart.
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 7
OR "robotics" OR "self-driving") AND ("government" OR "public man-
agement" OR "public sector" OR "public administration" OR "public pol-
icy") AND ("accountability" OR "accountable" OR “transparency” OR
“transparent”). The search terms comprise three components. The first
part encompasses keywords related to AI, which have been crafted based
on a prior systematic review on the use of AI in public governance
(Zuiderwijk et!al., 2021). The purpose of applying this list of search
words is to capture nearly all possible technologies related to AI. The
second part defines the public sector context. The third part aids in
identifying literature related to accountability. Notably, in response to
previous research highlighting transparency as an integral facet of account-
ability, we have also incorporated transparency into the search query
(Busuioc, 2021).
We applied these terms to search for peer-reviewed journal articles
written in English and published between January 2013 and July 2023.1
We searched the literature using international research databases, including
We b o f S c ie n ce ( Wo S) , E BS CO , a n d S c op us .2 The searches yielded 533
results (248 in Web of Science, 107 in EBSCO, 178 in Scopus). After
deleting duplicates, we obtained 385 unique studies.
As far as the phase of screening, we used three criteria to include
studies in the review: (1) Field: Articles should study the use of AI in the
public sector, which, according to the OECD, includes the general gov-
ernment sector plus all public corporations3 (OECD, 2014). (2) Topi c:
Articles need to adequately emphasize the accountability aspect. (3)
Methods: Both empirical and conceptual/theoretical studies are considered.
However, systematic reviews were excluded to avoid including duplicate
publications. The collected articles were screened in two steps. In the first
step, both authors screened the titles, abstracts, and keywords against the
eligibility criteria of the field and study design to perform initial filtering.
347 publications were removed after this stage of evaluation. Next, the
full text of the remaining articles was screened against the eligibility cri-
teria of topic to determine if the article provided details on the issues
concerning the accountability of AI-based systems. The articles were read
and assessed by both authors. The screening and eligibility evaluation
eventually led to the inclusion of 36 publications.
As for data analysis, a detailed metadata framework was designed to
categorize the included articles. The scheme contains (1) descriptive infor-
mation, including publication number, publication type, searching source,
publication year, name of authors, the title of the article and journal/
conference; (2) the purpose of research or research questions; (3) the theo-
retical approach or framework; (4) method-related information, which indi-
cates whether the paper is empirical, what particular method is utilized,
and which policy and geographic area the evidence belongs to; (5) findings
8 YUAN AND CHEN
and contributions. The data were coded manually by one of the authors
to ensure consistency (Saldaña, 2015); the coding results were then reviewed
and evaluated by the other researcher. All the literature data were coded
using MAXQDA 22.2.9 software.
Findings
Descriptive analysis
In terms of publication years (see Figure 2), although our literature search
spanned from 2013 to 2023, all the publications included in the analysis
were published after 2018. Furthermore, there was a noticeable surge in
the number of publications after 2020. Tab le 1 provides insights into the
journal sources of these publications. The most frequently occurring jour-
nals were Big Data and Society, Government Information Quarterly, Public
Administration Review, and Technology in Society, each contributing two
publications. Besides, our analysis demonstrates that research on the
accountability of AI-based systems in the public sector is highly interdis-
ciplinary, encompassing academic fields such as public administration, law,
information systems, digital government, and science, technology, and
society (STS) studies.
Regarding the type of study, as depicted in Figure 3, approximately half
of the publications (17 out of 36) consist of conceptual or literature reviews.
16 out of 36 publications employed a qualitative approach, predominantly
legal analysis and case studies. Only three publications adopted a quanti-
tative approach. In terms of policy area (Figure 4), among the publications
that explicitly mentioned the domains when discussing the application of
AI-based systems, social welfare and criminal justice (e.g., policing, sen-
tencing, surveillance, etc.) were the most frequently studied areas. As
Figure 2. Year of publications.
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 9
shown in Figure 5, most of the publications (26 articles) focus on the use
cases in Europe and North America, whereas very few studies discuss the
use of AI-based systems in other regions like Asia and Africa (6 articles).
Finally, the AI-based systems under examination in all the included studies
primarily served to support decision-making processes.
Table 1. Sources of the included publications.
Journal Number of publications
Big Data and Society 2
Government Information Quarterly 2
Public Administration Review 2
Technology in Society 2
Accounting Auditing & Accountability Journal 1
American Journal of Comparative Law 1
Asia Pacic Journal of Public Administration 1
Brigham Young University Law Review 1
Canadian Journal of Communication 1
Columbia Law Review 1
Computer Law and Security Review 1
Daedalus 1
Data & Policy 1
Ejournal of Edemocracy and Open Government 1
Ethics and Information Technology 1
European Journal of International Law 1
Human Rights Law Review 1
Information Development 1
Information Polity 1
International Journal of Electronic Government Research 1
International Journal of Law and Information Technology 1
International Journal of Public Administration in the
Digital Age
1
Journal of Social Policy 1
Journal of the Knowledge Economy 1
Law, Technology and Humans 1
Monash University Law Review 1
Social Science Computer Review 1
Sustainability (Switzerland) 1
Telecommunications Policy 1
Tilburg Law Review 1
University of Michigan Journal of Law Reform 1
Yale Journal on Regulation 1
Figure 3. Study type.
10 YUAN AND CHEN
Figure 6 shows the main themes identified in the current studies about
accountability of AI-based system in the public sector. The systematic review
has provided a more comprehensive understanding of key actors, mecha-
nisms, and enablers/challenges to hold AI-based systems accountable in the
public sector. Below, we will discuss the review findings on these main themes.
Key actors involved in the accountability of AI-based systems
According to Bovens (2007), several actors are involved as either account
takers (forums) or account givers (actors) in holding public organizations
accountable. The account givers are accountable to many different account
takers, all of which apply a different set of accountability criteria. In the
context of AI-based systems in the public sector, scholars have identified
Figure 4. Policy area.
Figure 5. Geographical distribution.
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 11
four key account takers to whom accountability must be rendered, and
three primary actors serving as account givers.
Account takers
Political accountability.!With the increasing use of AI-based systems in public
sector decision-making, recent studies have begun to treat the incorporation
of those systems as a governance choice. As a result, there is growing
advocacy for greater recognition and scrutiny of the rationale and justi"cation
behind such choices, as well as a more comprehensive understanding of their
impact on fundamental principles and established models of accountability
(McGregor, 2018). In this regard, political overseers play a crucial role in
providing external oversight of government organizations’ use of AI-based
systems in decision-making (Bignami, 2022). ey are viewed as regulators
responsible for examining the model setup of AI-based systems. Scholars
argue that while AI-bases systems are oen presented as “neutral devices,”
they might inherently encode competing values or value tradeos within
their algorithms, necessitating political deliberation on how to balance these
elements (Bracci, 2023; Considine et!al., 2022). erefore, processes and
decisions related to AI governance and policy must be subject to continuous
scrutiny by political overseers. Political actors must remain vigilant to these
considerations and actively participate in discussions and deliberations on AI
design and operation (Busuioc, 2021).
Another commonly mentioned group of actors is the general public or
individual citizens (Selten & Meijer, 2021). It is crucial that citizens develop
a broad understanding of an algorithm’s strengths and limitations to over-
come fear of the unknown. Their perspectives play a critical role in
Figure 6. Main themes identied in the literature.
12 YUAN AND CHEN
identifying, interpreting, and addressing the social ramifications of AI algo-
rithms in practice (Buhmann & Fieseler, 2021). For individual citizens,
scholars argue that they are entitled to access information about how these
systems function and how their rights are affected so as to ensure algorith-
mic accountability (Rachovitsa & Johann, 2022). This access enables citizens
to learn about, validate, and correct errors in algorithmic decision making,
especially if AI-based systems violate their rights (Engstrom & Ho, 2020;
Liu et!al., 2019). Also, it is noted that mass media can act as a channel to
link between the public and government organizations in holding the use
of AI-based systems accountable (Buhmann & Fieseler, 2021).
Legal accountability.!Some scholars mentioned judicial judges as a key actor
to ensure legal accountability within the courts. ey provide both ex ante
and ex post review of AI-based systems adopted in the public sector.
Crawford and Schultz (2019) argue that judicial judges shall provide ex
ante scrutiny of vendors who supply AI-based systems for government
decision-making. Given that the use of AI-based systems by government
agencies may lead to litigation in various circumstances, judges also play a
crucial role in ex post oversight (Bignami, 2022, p. 16). Upon citizens’
appeals, judicial courts conduct reviews of agency decisions using AI-based
systems to determine whether they undermine human rights and the rule
of law (Engstrom & Ho, 2020). is judicial review goes beyond merely
assessing the decision itself; it evaluates the entire decision-making process,
considering various elements both individually and collectively to ensure
adherence to principles of good decision-making under administrative law
(Cobbe & Singh, 2020).
Administrative accountability.!It is suggested that government agencies’ AI-
based systems should be subject to appropriate and regular internal oversight
(Bignami, 2022, p. 16). is internal oversight comes from a variety of
government sources, such as inspectors general, externally facing ombuds,
or the Government Accountability O&ce. Government agencies are
recommended to establish a protocol for regularly evaluating AI-based
systems throughout their lifespans. Some scholars also suggest creating
specialized audit bodies for AI ethics, standardization, and regulation
compliance, tasked with monitoring the design and operation of AI-based
systems (Busuioc, 2021). In addition, it is proposed that such regulatory
bodies should have an appeal function, allowing citizens dissatis"ed with a
decision, its explanation, or in cases of disputes, to appeal the decision to
the agency for arbitration (Bannister & Connolly, 2020).
Professional accountability.!Another important account takers mentioned in
the literature are domain professionals or experts. Expert auditing of
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 13
algorithms is essential to ensure they function correctly and comply with
legal standards (Selten & Meijer, 2021). Given their specialized knowledge
of AI model setup and design, experts are critical in identifying potential
aws in algorithms, such as coding errors or biases. Professional
accountability is particularly important because lay audiences—including
individuals adversely aected by algorithmic decisions—oen lack the
expertise required to understand the features and operation of algorithms,
leaving them unable to meaningfully contest or challenge these decisions or
seek redress (Busuioc, 2021).
Account givers
Most scholars suggest that AI accountability needs to be considered
throughout the different phases of the AI life cycle in public service, where
multiple human actors, or “responsibility points,” are obligated to explain
and justify their actions (Bannister & Connolly, 2020; Bracci, 2023). In
this context, scholars identify both governmental and non-governmental
actors as crucial for ensuring AI accountability. The literature predomi-
nantly highlights three types of actors in this regard.
AI developers or companies.!As far as non-governmental actors, scholars
mentioned that private IT companies or AI developers should be held
accountable. Public organizations oen outsource AI-based system
development and/or implementation to private IT companies. ese private
companies, frequently the designers, developers, and maintainers of AI
systems, are thus viewed as responsible for their “arti"cial technocrats” (Sætra,
2020). With their technological expertise, they are considered the most
capable and cost-eective parties to assume accountability in the development
and use of AI-based systems (Bell et!al., 2023). Given their direct inuence
on government decisions, private IT companies should be treated as state
actors, bound by public laws. Special measures should be implemented in
contracts and terms that enforce greater accountability (Liu et!al., 2019).
Decision makers in adopting AI.!For governmental actors, scholars "nd that
decision makers who decide to use of AI-based systems in public organizations
are one of the most important actors. An AI developer cannot be expected
to fully grasp the nuances of how and where to use AI-based systems within
government settings. Decision-makers typically de"ne the scope and inuence
of AI algorithms in the service delivery process, determining how and where
these algorithms are deployed. eir role is critical because, through their
decisions, they not only de"ne the system’s role but also impart values in the
design and implementation of AI-based systems (Bracci, 2023). Decision
makers, therefore, need to be held accountable for their decisions and the
resulting consequences (Bracci, 2023).
14 YUAN AND CHEN
Users of AI-based systems.!Another group of governmental actors is the user
of AI-based systems, such as individual frontline civil servants who use
these systems in their administrative tasks (Cobbe & Singh, 2020; Contini,
2020). In the decision-making process, these users are oen regarded as
having superior authority over the technology, particularly in various parts
of the decision-making process. ey remain the "nal decision-makers,
determining whether applicants receive approval. ese users decide whether
to use the results generated by AI-based systems and to what extent those
results inuence their "nal decisions. AI-based systems thus play a secondary
or supportive role in the process (Ranerup & Henriksen, 2019). In this
sense, the “human in the loop” or “human on the loop” should be seen as
key individuals accountable for "nal decisions, which will be further
elaborated on in the subsequent section (Carney, 2020; Parycek et!al.,
2023).
Figure 7 illustrates the accountability relationships discussed in the
literature. Most studies emphasize the connections between decision-makers
to adopt AI and the four distinct types of account takers. Besides, AI-based
system users are highlighted as needing to provide accountability to both
the citizens impacted by their decisions and the administrative or expert
audit units, particularly if a decision appears to infringe on citizens’ rights.
Regarding AI developers, the literature suggests that they should be
accountable to the administrative or expert audit units, the judicial courts,
and political overseers.
Mechanisms to hold AI-based systems accountable
Following Bovens (2007)’s framework on accountability, our analysis
shows that current literature has discussed mechanisms on three dimen-
sions of accountability: information, explanation/justification, and judg-
ment/consequence. Most studies focus on the mechanisms that promote
Figure 7. Accountability relations between account givers and account takers.
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 15
information and explanation on their design and operation, as one of
the primary challenges to accountability lies in their high level of opacity.
Some research has begun to explore mechanisms that ensure the trans-
parency and explainability of AI influence on human decision-making,
recognizing that many decisions are algorithmically informed. However,
the dimensions of judgment and consequence have received relatively
limited attention.
Mechanisms for information
Opening the black box of algorithms to understand their functioning and
identify any biases related to inputs and/or outputs is crucial for holding
AI-based systems accountable (Bracci, 2023; Cobbe & Singh, 2020). The
literature emphasizes that transparency in AI accountability hinges on
providing information about data inputs, factors/parameters in the models,
and the computation/program processes (Liu et!al., 2019; Plantinga, 2024).
Since algorithms rely on input data, visibility into training data allows
account takers to evaluate to what extent the data is discriminatory, inac-
curate, or biased (Kuziemski & Misuraca, 2020). Multiple aspects of train-
ing data, such as sources of data, data ethics, data filtering process, and/
or data synthetic process, are highlighted as essential for ensuring trans-
parency (Henman, 2020; Parycek et!al., 2023).
Other scholars argue that accountable AI-based systems must also pro-
vide information on how input data is translated into decisions (König &
We nz e lb u rg e r, 2020). This includes the factors, parameters, and computa-
tional processes used in the algorithms that contribute to predictions
(Rachovitsa & Johann, 2022). In machine learning algorithms, particularly
those involving neural networks, this often involves multiple hidden layers
of artificial neurons, with thousands of features and millions of weights
connecting inputs to outputs (König & Wenzelburger, 2020). The process
also involves numerous iterations of data analysis or simulations that lead
to model construction (Redden, 2020). Unpacking the internal “features”
of AI-based systems is essential for understanding the logic of algorithms
and assessing whether these systems comply with legal and ethical require-
ments (Brand, 2022) and reflect the complexity of reality (Jørgensen &
Nissen, 2022).
Mechanisms for explanation/justication
Despite achieving transparency or accessibility within AI-based systems,
accountability may still not be fully ensured due to the lack of explanation
for algorithmic results and justification for their use (Busuioc, 2021). The
literature highlights the importance of explaining the rationale behind
specific decisions informed by AI-based systems (Bracci, 2023). This entails
16 YUAN AND CHEN
providing explanations for both the outcomes generated by AI-based sys-
tems and the human decisions influenced by these outcomes. Besides, as
mentioned previously, some scholars recently emphasized that the adoption
of AI-based systems should be viewed as a governance choice, underscoring
the importance of explaining the decision to adopt AI in the first place.
First, most scholars focus on the explainability or interpretability of
AI-based systems and propose a tactic known as explainable AI, which
emphasizes that both the processes and outcomes should be accessible
and understandable to citizens (Chen et!al., 2023). Explainable AI may
involve creating an explainer algorithm that can independently provide an
approximate explanation of how an AI system generated a particular
decision (Janssen et!al., 2022) or offering general reports on the logic and
impact of the AI-based system (König & Wenzelburger, 2020). Although
these explanations may not be exact, they highlight key parameters within
the AI and the value tradeoffs inherent in the model, enabling users to
assess the system against criteria of accountability (Buhmann & Fieseler,
2021). Similarly, “adversarial testing” involves attempting to “break” an
AI-based system or induce it to make incorrect decisions, which helps to
identify critical parameters that lead to the system’s outcomes and allows
for the evaluation of its logic and accountability in specific cases (Henman,
2020). Studies suggest that explainable AI significantly enhances trust in
AI systems (Grimmelikhuijsen, 2023) and aids experienced decision-makers
in detecting incorrect decisions (Janssen et!al., 2022).
Second, since most AI-based systems currently play a supportive role
for human decision-makers, scholars argue that it is crucial to scrutinize
every decision and each stage of the process leading to the final decision,
rather than focusing narrowly on the algorithms or models themselves
(Bannister & Connolly, 2020; Cobbe & Singh, 2020). In this sense, users
of AI-based systems remain accountable for the consequences of their
decisions to use results of AI (Contini, 2020). In the literature, users’
explanation of how they use the results are found critical to accountability
in two aspects. On the one hand, most studies emphasize the importance
of keeping a human in the loop, suggesting that human justification is
necessary to prevent overreliance on AI-based system recommendations
and to correct potential algorithmic unfairness or discrimination (Carney,
2020; König & Wenzelburger, 2020). Human decision-makers’ knowledge
and ethical concerns serve as safeguards that legitimize algorithmic deci-
sion-making systems (Bignami, 2022; Waldman & Martin, 2022).
Additionally, detecting human bias is crucial to ensuring that decision-mak-
ers act as genuinely independent evaluators of AI outputs, rather than
merely rubber-stamping them (Parycek et!al., 2023). On the other hand,
recent studies indicate that the interaction between AI-based systems and
their users can gradually enhance the systems’ accuracy in making
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 17
decisions and their accountability (Jørgensen & Nissen, 2022). Human
explanations, which capture the complexity and ambiguity of real-world
situations, contribute to the development of more adequately reasoned
justifications on individual cases within AI-based systems (Carney, 2020),
which, in turn, enhances the accuracy of AI results and ensures greater
accountability.
Third, rather than focusing solely on opening the black box of AI-based
systems and examining the accountability of specific results or decisions,
scholars have recently begun to consider the use of AI-based systems as
an assemblage of human and non-human actors, where an agency between
data and people is created within a particular institutional and cultural
context (Bracci, 2023). Therefore, the adoption of AI-based systems is
viewed as a governance choice that warrants scrutiny. Approaching AI
from the perspective of governance choices, rather than merely assessing
its influence on specific decisions, provides a broader understanding of
its impact on bureaucratic discretion and whether changes to that discre-
tion are legitimate and acceptable (McGregor, 2018). This allows evaluation
on the full mandate of the actor rather than only on the direct effects of
the technology concerned.
Mechanisms for judgment/consequence
While most studies focus on the previous two dimensions of accountability
to address inherent opacity in the AI-based systems, the third dimension,
judgment and consequences, have received limited attention in the liter-
ature. Some studies that do touch on this dimension propose AI impact
assessment as a major tool for making judgments. However, mechanisms
for rectification and assigning consequences to specific account-givers
remain underexplored.
Scholars argue that impact assessment of AI-based systems is needed
before introduction of those systems to ensure their accountability (Brand,
2022; Rachovitsa & Johann, 2022). For instance, the UK government has
issued guidelines focused on assessing, planning, and managing AI-based
systems, with an emphasis on using AI ethically and safely (Henman,
2020). Similarly, the Government of Canada proposed a questionnaire to
evaluate potential impacts on individuals’ or communities’ rights, health,
or well-being, economic interests, and the sustainability of ecosystems
(Kuziemski & Misuraca, 2020). To enable such evaluations, it is recom-
mended to provide more information about where and how these systems
are being implemented, make greater efforts to generate public deliberation
about their use, and conduct more in-depth investigations into their impact
on practitioners (Redden, 2020). Still, such assessment cannot be conducted
without first making accountability criteria explicit (Bracci, 2023).
Challenges remain in distinguishing between discrimination or bias and
18 YUAN AND CHEN
personalization or customization (Henman, 2020). It is essential to establish
standard of fairness, ethical, unbiased algorithms to resolve tensions and
tradeoffs among value notions advocated by different actors, a process
that requires a priori deliberation and interrogation of what is acceptable
(Busuioc, 2021).
In this sense, scholars propose a governance approach involving mul-
tiple stakeholders in the assessment of AI accountability. Multi-actor
involvement and deliberation are essential for the timely identification
of sensitive issues, concerns, or value-laden biases, as well as for estab-
lishing acceptable criteria (Franzke et!al., 2021; Redden, 2020). Scholars
argue that this participatory approach is not only important for assessing
accountability after the implementation of AI-based systems but is also
critical in involving stakeholders early in the design process to ensure
transparency, explainability, and accountability are integrated from the
outset (Bell et!al., 2023; van den Homberg et!al., 2020). Buhmann and
Fieseler (2021) develop a deliberative framework for responsible innova-
tion in AI, which includes AI developers, companies, civil society, and
media actors to facilitate broader public debate and discourse, contrib-
uting to responsible innovation processes. Others suggest a user-centered
design approach or bottom-up approach that integrates users’ professional
knowledge and public opinion into the design process and continuously
monitors interactions to identify negative effects at an early stage and
to make design adjustments (Bloch-Wehba, 2022; Parycek et!al., 2023).
To ensure the effectiveness of such deliberative engagement in finding
optimal solutions, this process should satisfy conditions of inclusiveness
to various stakeholders, ensure the reciprocity of actors, and facilitate
diverse and well-informed arguments and viewpoints (Buhmann &
Fieseler, 2021).
Enablers and challenges to hold AI-based systems accountable
Scholars have explored the enablers and challenges of the aforementioned
mechanisms from technological, organizational, and institutional perspec-
tives to ensure their effective implementation. Much of this research has
concentrated on identifying challenges in ensuring the adequacy of specific
accountability arrangements or comprehensive accountability regimes
designed to effectively govern the use of AI-based systems within the
public sector. Following Boven’s framework, this section presents the iden-
tified enablers and challenges.
Information-related
While unpacking the internal features of algorithms and data inputs is
highly desirable, opening the black box of AI-based systems continues to
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 19
face significant challenges. Technologically, as algorithms become increas-
ingly detached from human control, it becomes more challenging to explain
and interpret how AI-based systems arrive at specific results. This is
particularly true for deep learning algorithms or neural networks, where
decisions are the result of complex processes involving synthetic data
production, with hundreds of layers, features, and weights contributing to
a single outcome (Busuioc, 2021). Also, software codes in these systems
often undergo iterative changes, which complicates the task of describing
these algorithms compared to traditional technologies (Bracci, 2023; Carney,
2020). In this sense, information regarding those algorithms at one time
point may not fully reflect the complexity of AI-based systems overtime.
To a dd re s s t h es e i ss ue s , B r an d ( 2022) proposes a sandbox approach that
tests new AI applications throughout their entire lifecycle.
Organizationally, it is argued that account takers may lack the necessary
organizational capacity to effectively conduct human inspections of AI-based
systems. Challenges in data quality and data management practices within
public organizations are noted as potential barriers, limiting their ability
to identify racial biases and discrimination in datasets (Franzke et!al.,
2021). These limitations can subsequently lead to biased outcomes in
AI-based systems (Richardson & Kak, 2022). When AI-based systems are
owned by private companies, access to their code or inner technical func-
tioning can be limited or even impossible due to trade secret protections.
This reduces the incentives of private companies as account givers to
provide sufficient information for accountability in AI-based systems
(Busuioc, 2021).
Institutionally, data protection regulations may restrict access to sensitive
data, hindering the ability to review the information used by AI-based
systems. This challenge is exacerbated by ethical and privacy concerns,
particularly when scrutiny involves disadvantaged populations (Franzke
et!al., 2021). Additionally, full disclosure of software and data sources
poses challenges, as it could expose vulnerabilities that might be exploited.
Moreover, some algorithms remain inaccessible because governments are
concerned that citizens subjected to these algorithms might “game” the
system if they fully understand its operation (Contini, 2020;
Grimmelikhuijsen, 2023). These institutional issues will significantly influ-
ence whether, and to what extent, information is accessible to the
account-takers.
Explanation/justication-related
At the organizational level, while the human-in-the-loop approach is vital
for providing explanations of decisions informed by AI-based systems,
insufficient knowledge or expertise about these systems among public
employees can hinder their ability to effectively contest or challenge
20 YUAN AND CHEN
AI-generated decisions (Parycek et!al., 2023). In such cases, the human-
in-the-loop approach risks becoming a superficial check, where the medi-
ator lacks sufficient understanding of the system’s functioning—or
malfunctioning—thereby undermining meaningful oversight (Bracci, 2023).
Therefore, it is essential that public employees are empowered with the
necessary knowledge and confidence in their decision-making abilities to
serve as genuinely independent skeptics of AI-generated outcomes.
(Carney, 2020).
Another challenge is the tendency to render human judgment to
AI-based system, due to a general belief that these new systems are objec-
tive and neutral, leads to a normalization of not interrogating their outputs
(Redden, 2020). This can result in individuals being subjected to inaccurate
or unnecessary profiling and control (Jørgensen & Nissen, 2022). Public
employees might develop a reliance on these systems to the extent that
they no longer question the outputs, either due to ignorance or automatic
acceptance (Bracci, 2023). Frontline employees may come to prioritize the
use of AI-based systems as their primary method of working, basing
decisions on system outputs rather than their own professional experience
(McGregor, 2018).
Judgment/consequence-related
Most of the identified challenges are concentrated at the institutional level.
Scholars argue that the divergent values that AI-based systems are expected
to pursue make it challenging to hold the use of AI-based systems account-
able. Accountability criteria and standards are not always clear-cut
(Kuziemski & Misuraca, 2020), particularly because AI ethics are frequently
framed as vague standards or norms with limited articulation of their
underlying values (Bloch-Wehba, 2022). Accordingly, the practice of public
accountability involves complex tradeoffs among contending goals, claims,
and normative criteria (Busuioc, 2021). The variety of value positions held
by different stakeholders often obscures the core values that are needed
for effectively holding the use of AI-based systems accountable. For
instance, the risk concerns possessed by the society as a whole might be
different from the affected populations (Rachovitsa & Johann, 2022).
While scholars have advocated for a participatory and deliberative
approach to assess the accountability of AI-based systems, questions remain
about whether all actors have the ability to effectively express their opin-
ions and whether their viewpoints can be fully integrated into the deci-
sion-making process. Ordinary citizens’ ability to provide input and
feedback depends not only on their willingness to engage and reciprocate
but also on their capacity to offer informed suggestions in the complex
domain of AI (Buhmann & Fieseler, 2021). Many citizens lack the expertise
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 21
to question or challenge the findings of AI-based systems and to contest
decisions informed by these systems (Crawford & Schultz, 2019; Richardson
& Kak, 2022). The effectiveness of citizen involvement in fostering deep
discursive engagement depends on adherence to normative principles such
as participation, comprehensibility, multivocality, and responsiveness
(Buhmann & Fieseler, 2021).
Lastly, the absence of clear rules and regulations governing the use of
AI-based systems in the public sector poses significant challenges to making
judgments or determining appropriate consequences. While some policy
documents address the ethical and legal issues of automated decision-making,
they often fall short of proposing practical measures to prevent the misuse
of AI-based systems (Kuziemski & Misuraca, 2020). This highlights the need
for greater legal certainty regarding the accountable use of AI-based systems
(Brand, 2022; Busuioc, 2021). Moreover, several scholars emphasize the
importance of further exploring whether and how public law norms apply
to private IT companies and their employees involved in designing, imple-
menting, and maintaining these systems (Crawford & Schultz, 2019).
Discussion
This paper presents a systematic literature review focusing on the account-
ability of using AI-based systems in the public sector. Analysis of 36
publications indicates that this research area is highly interdisciplinary,
that most studies utilize conceptual analysis and qualitative methods, and
that human actors are perceived as the primary agents to hold accountable.
Additionally, we identify several mechanisms that support the accountability
of AI-based systems, particularly in terms of informing conduct and facil-
itating explanation. However, we also highlight several challenges that
could undermine these accountability mechanisms.
Regarding actor-forum relationships, our findings indicate that most
studies perceive human actors as those who should be held accountable
when AI-based systems are used for decision-making in the public sector.
It implies that AI-based systems in the public sector currently are not
perceived as equivalent as human intelligence. The responsibility for AI
adoption and use in the public sector necessarily lies with human actors.
Following Bovens (2007)’s framework and similar to previous studies about
accountability of using IT systems in the public sector, accountability in
AI-based systems is viewed from a relational perspective that implies
accounting for human conduct to various forums in a variety of ways.
Our analysis underscores the significant role of users as potentially crucial
checkpoints for enhancing transparency and explainability in the use of
such systems (Contini, 2020), while the developers and decision-makers
to adopt the systems also remain as significant account givers.
22 YUAN AND CHEN
Private IT companies and developers, previously underemphasized, have
also emerged as a critical category of actors responsible for designing and
training AI algorithms. As reliance on AI-based systems grows within the
public sector, these developers increasingly assume the roles of technocrats
within machine bureaucracies (Considine et!al., 2022). The involvement of
private companies in the use of AI-based systems has become so integral to
decision-making processes affecting the general public that their role could
be likened to that of public employees; therefore, they should adhere to public
laws (Liu et!al., 2019). The increasing adoption of these systems by govern-
ments, developed by private companies, suggests a complex interplay between
the public and private sectors, implying a shared responsibility and account-
ability for the outcomes of AI-based systems in decision-making processes.
For individual users, we identified a dual expectation: they are seen as
bearing both responsibility and accountability when using AI-based sys-
tems. As these systems, acting as agents for human users, transform dis-
cretion into a veneer of technical requirements—potentially obscuring
complex ethical decisions—users’ intrinsic professionalism and moral obli-
gation are seen as the primary safeguards against the digitalized discretion
inherent in these systems. This perspective aligns with Friedrich (1940)’s
view on administrative responsibility. However, users’ discretion or control
over AI-based systems must also be accountable to external actors, includ-
ing political overseers, judicial courts, and citizens, to ensure true account-
ability. This necessity arises because human users’ discretion and scrutiny,
grounded in their internal sense of responsibility, may not fully reflect
public interests due to potential biases or a lack of intent or capability to
critically assess the systems (Finer, 1941; Parycek et!al., 2023). These
findings suggest a balanced approach that integrates both internal respon-
sibility and external oversight.
Our review also identified key mechanisms—information, explanation/
justification, and judgment/consequence—for holding the use of AI-based
systems accountable and the associated enablers and challenges. Regarding
the information, explanation, and justification components, the findings first
highlight that the AI-based systems are not merely technological tools applied
in the public decision-making process but rather an assemblage of human
and non-human actors, collectively creating agency within specific institu-
tional and cultural contexts (Bracci, 2023; Richardson & Kak, 2022).
Consequently, simply unveiling the features of AI-based systems is insuffi-
cient. Equally important—if not more so—is explaining the dynamics of
human-AI interaction to gain a comprehensive understanding of the factors
influencing human decisions utilize AI-based systems and their outcomes.
Second, while the black-box nature of AI is often considered a persistent
challenge in public administration and law literature, only a few scholars
have acknowledged the advancements in Explainable AI (XAI). In contrast,
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 23
scholars in other disciplines have been actively exploring technical
approaches to further enhance the explainability of AI, particularly through
development on XAI. This body of work emphasizes three levels of expla-
nation: data explainability, model explainability, and post-hoc explainability
(Ali et!al., 2023). Researchers have been developing and assessing various
applicable techniques to improve the explainability of AI algorithms across
these dimensions (Dwivedi et!al., 2023; Fernández-Loría et!al., 2022; Hassija
et!al., 2024). These efforts aim to provide two types of explanations: global
explanations, which explain the functioning of the overall machine learning
model, and local explanations, which clarify the relationship between
specific input-output pairs or the reasoning behind results (Mohseni et!al.,
2021). These advancements could significantly enhance the use of AI-based
systems in governmental contexts by improving users’ understanding of
how these systems function. However, some scholars argue that explainable
AI may face a tradeoff between explainability (ease of interpretation) and
accuracy (the complexity of AI models), potentially failing to fully capture
the logic or impact of AI-based systems (Bell et!al., 2023). Besides, there
are concerns that small changes in input data or model parameters can
produce significantly different explanations, raising questions about the
reliability of these explanations (Abusitta et!al., 2024). Future research is
needed to further investigate the implications of XAI in promoting account-
ability in the use of AI-based systems.
Regarding the judgment/consequence component, our analysis indicates
that AI accountability is co-created through the collaboration of multiple
actors. Emphasis is placed on ensuring accountability throughout the entire
lifecycle of AI-based systems, rather than focusing solely on the imple-
mentation phase. Involvement of multiple stakeholders at the beginning
of the development allows AI-based systems to be designed with account-
ability as a built-in feature. As those systems contain not only the design
of technological functioning but also incorporation of competing value
propositions or tradeoffs, involvement of multiple actors from both the
public and private sectors seems essential. The relations between account
takers and account givers seem to shift toward a proactive communication
and deliberation on the meaning of transparency and accountability
(Bracci, 2023).
Our results also show that, like other IT systems in the public sector,
accountability of AI-based systems is shaped by a combination of socio-tech-
nical enablers and challenges. Despite the complexity in the AI-based
systems, the use of such systems is still confronted with the classic “many
eyes and many hands” problem in accountability (Bovens, 2007). In the
lack of a resolution to tradeoffs among divergent public value propositions,
the ambiguity in operationalization and standards of accountability places
challenges for developers to translate them into practical instructions and
24 YUAN AND CHEN
encode them into AI-based systems. To make things worse, the distinction
between human discourse and programing codes may lead to an unfaithful
translation of the value propositions to algorithms. The technological and
propriety black box further reduce account takers’ ability to examine the
extent to which algorithms reflect public interests. The accountability of
AI-based systems needs to be re-considered in terms of how they are
enacted in decision-making process and their impact on human discretion
to determine legitimate and acceptable interaction reflecting public interests.
The review findings can also be linked to the broader literature on
government accountability. In this regard, our review highlights various
conceptualizations of accountability. When considering accountability as
either sanction-oriented or trust-oriented (Mansbridge, 2014), several stud-
ies, particularly those from the field of law, emphasize sanction-based
accountability. These studies tend to express distrust toward the systems,
their developers, and the associated government agencies. Conversely,
studies mainly from the field of public administration adopt a trust-based
perspective, advocating for collaboration among various stakeholders to
ensure AI use meets accountability standards. Furthermore, we identify
an interlinked view of outcome- and process-based accountability (Patil
et!al., 2014) in the context of AI use in government. The need to hold
AI systems accountable appears to stem from concerns about whether the
systems produce democratic outcomes and the processes by which specific
decisions are made. Therefore, both the process and the outcomes related
to AI use should be thoroughly scrutinized.
Finally, several notable similarities can be observed when comparing
the findings to the literature in other fields, such as Computer Science
and Information Systems. Both streams of literature recognize the potential
harms that AI poses to individuals and society, emphasizing the critical
importance of ensuring accountability in the use of AI-based systems (Li
et!al., 2023; Rahwan et!al., 2019; Saeed & Omlin, 2023). In addition, both
fields view human stakeholders as ultimately responsible for any harm
caused by these systems and identify multiple stakeholders from society,
government, and industry as key gatekeepers in ensuring accountability
(Rahwan et!al., 2019). Moreover, both literatures adopt a life-cycle per-
spective on AI-based systems, stressing the necessity of implementing
accountability measures throughout the stages of design, development, and
deployment (Li et!al., 2023; Saeed & Omlin, 2023).
Research agenda
Based on our analysis of the literature and comparison with other fields,
we offer several avenues for future research in order to deepen our under-
standing of the accountability of AI-based systems within the public sector
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 25
across various levels. First, at the individual level, our review highlights
the significant roles of human users in ensuring the accountable use of
AI-based systems, underscoring the need for further research on the capa-
bilities required for users to fulfill these roles effectively. This line of inquiry
could be informed by existing research on bureaucratic behavior, partic-
ularly in how bureaucrats navigate responsibility and accountability
demands (Su & Wang, 2024). Additionally, from a human resources per-
spective, it is crucial to explore the technological understanding and train-
ing necessary for users to meet accountability requirements. Furthermore,
insights from Human-Computer Interaction (HCI) research could be valu-
able in examining how users interact with AI-based systems when making
judgments and decisions (Jung & Camarena, 2024). Ultimately, advancing
this research agenda will contribute to a more comprehensive understand-
ing of how to empower human users in their critical role of upholding
the accountability of AI-based systems.
Second, at the organization level, our understanding of accountability
dynamics could be enhanced by examining these dynamics within diverse
organizational contexts. The studies in our review seldom distinguish the
use of AI-based systems in different organizational settings. However,
previous research has shown that the impact of AI-based systems on public
organizations and employees may vary based on organizational character-
istics and administrative culture. For example, Bullock et!al. (2020) high-
lights that the relative effectiveness of human versus AI input can differ
depending on the complexity (i.e., level of deviation from the norm) and
uncertainty (i.e., level of analyzability) of a task faced by an organization.
Additionally, Young et!al. (2019) argue that AI can influence organizations
by affecting the availability and quality of information, as well as organi-
zational attitudes and values. Further, the interaction between public
employees and AI-based systems is also influenced by organizational cul-
ture. A culture that encourages professional judgment in public employees,
as opposed to one that prioritizes adherence to hierarchy and control,
may help mitigate potential algorithmic biases (Meijer et!al., 2021). These
insights reveal that organizational characteristics and administrative cultures
can lead to differing requirements in the design and use of AI-based
systems. Further research is warranted to explore these nuances across
various organizational contexts.
Third, at the technological level, more effort can be made to uncovering
the potential heterogeneity in accountability dynamics among different appli-
cations of AI technologies. Our literature review reveals that discussions
on AI accountability predominantly concentrate on AI-based systems used
for decision-making support, primarily involving machine learning tech-
nologies. Notably, previous research highlights Artificial Neural Networks
(ANN) as the most frequently mentioned AI technique in studies
26 YUAN AND CHEN
examining AI-based systems for forecasting and decision-making support
in the public sector, especially in environmental protection and housing
sectors (de Sousa et!al., 2019). However, other studies have identified
alternative applications of AI-based systems, such as chatbots for customer
service (e.g., Chen et!al., 2024; Chen & Gasco-Hernandez, 2024; van
Noordt & Misuraca, 2022; Vereinte Nationen, 2022). The use of AI in
decision-making support is associated with higher risks and concerns about
diminishing human discretion and algorithmic bias (Bullock, 2019; Henman,
2020). In contrast, retrieval-based chatbots functioning as virtual assistants
pose lower risks (Chen & Gasco-Hernandez, 2024; van Noordt & Misuraca,
2022). These diverse applications of AI-based systems suggest varying
relationships between account givers and takers. Therefore, both conceptual
and empirical research should be directed toward exploring the similarities
and differences in accountability dynamics among distinct types of AI
technologies.
Fourth, more research can be conducted to examine the role of conse-
quences in the accountability dynamics. Our findings reveal that, in com-
parison to other elements depicted in Figure 6, the aspect of consequences
has garnered less attention from scholars. Admittedly, multiple studies
suggest that informing, debating, and judging are essential prerequisites
for determining consequences; therefore, it seems more crucial to identify
and address existing gaps in these preliminary processes (Bracci, 2023;
Busuioc, 2021). However, previous research in the field of government
accountability highlights the significance of the design of consequences,
including both rewards and sanctions (Mansbridge, 2014). These elements
are critical as they serve as incentives influencing the willingness of actors
to be fully accountable. Therefore, we advocate for further conceptual and
empirical research to elucidate the relationships between consequences and
the behavior of the actors and forums.
Fifth, more empirical evidence is needed to investigate how each element
in the model and their interconnections present in practice. Our analysis
demonstrates that most studies predominantly focus on diagnosing account-
ability gaps associated with the use of AI-based systems in the public
sector and offer recommendations for bridging these gaps. Consequently,
the elements and their interrelationships as illustrated in Figure 6 and 7
can be viewed as an idealized conceptual model. This model is deemed
crucial by researchers for ensuring accountable AI deployment in public
services. However, it remains uncertain how, and to what extent, each
element of this model and their interconnections are reflected in real-world
scenarios. To address this gap, additional empirical research is needed to
investigate the practical implementation of each component and their
interrelations in AI use cases. For instance, conducting case studies could
provide insights into the involvement and interplay of different actors and
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 27
forums in specific instances. Surveys and interviews also could be useful
tools for evaluating how various enablers and challenges influence the
dynamics of accountability in these contexts.
Finally, reflecting on our literature comparison in the previous section,
exploring the impact of Explainable AI development on government account-
ability emerges as a valuable and intriguing avenue for future research.
Current scholarship in this area predominantly focuses on systemic archi-
tecture in general or on applications within the business and medical
sectors, leaving a significant gap in our understanding of its implications
for public administration. Specifically, little is known about how govern-
ment entities, acting as technology buyers rather than regulators, can assess
the extent to which AI systems are explainable and suitable for use.
Furthermore, as Explainable AI has the potential to address current con-
cerns over the lack of transparency and explanation, it becomes essential
to reevaluate the accountability challenges associated with the deployment
of AI-based systems in the public sector. Addressing these gaps could
advance our understanding of how Explainable AI intersects with govern-
ment accountability, potentially reshaping the conceptualization of trans-
parency and explanation in the governance of AI-based systems in
government.
Conclusion
The use of AI-based systems for decision-making in the public sector is
an increasingly heated topic for both academia and practitioners. It brings
potential benefits to strengthen quality of decision making but also leads
to gaps in the chain of accountability. Through our systematic review, we
have delineated the current state-of-the-art scholarly perspectives on key
aspects of this topic. We identified the primary stakeholders involved in
the accountability process of AI-based systems, explored various mecha-
nisms that can bolster accountability, and examined factors that either
facilitate or hinder these mechanisms. Our review has uncovered several
gaps within the existing literature. To address these gaps, we propose
various directions for future research, which will provide deeper, more
nuanced insights that will be valuable for future researchers and practi-
tioners in this field.
Our study contributes to the current scholarship in several ways. We
synthesized the existing knowledge on accountability in the use of
AI-based systems within the public sector, mapping out our findings
using the accountability framework in public administration. This approach
provides a comprehensive perspective on the key questions of who, what,
when, and how in relation to the accountability of AI-based systems.
Additionally, by comparing our findings in the fields of public
28 YUAN AND CHEN
administration and law with the literature from other disciplines, we offer
valuable insights into the similarities and differences in how scholars
across fields conceptualize the key ideas of explanation and accountability
of AI-based systems. Particularly, our examination of recent developments
in Explainable AI introduces a new angle for considering the future roles
of bureaucrats and governments in deploying and managing AI-based
systems.
Our findings have significant implications for practitioners. The iden-
tified actors, their roles, and their interrelationships offer policymakers
and public employees valuable insights into the critical checkpoints that
can help ensure the accountable use of AI-based systems. Besides, we
encourage public organizations to remain mindful of the key mechanisms
necessary for maintaining accountability in AI usage and to proactively
address challenges that might hinder these mechanisms from functioning
effectively. Specifically, we emphasize the importance of aligning AI systems
with intended public values, improving the explainability of both the sys-
tems and the decisions, and enhancing the awareness, knowledge, and
capabilities of system users.
We acknowledge certain limitations arising from the choices made in
our review methodology. First, while we endeavored to use a compre-
hensive list of keywords in our literature search, we recognize that this
list may not be exhaustive, and as a result, some relevant articles may
have been missed. Second, although we also found literature written in
non-English languages, we ultimately included only English publications
due to our inability to read these other languages. Consequently, we may
have overlooked different perspectives on the accountability of AI-based
systems presented in these publications. Third, our inquiry is focused
solely on the use of AI-based systems within the public sector, which
means we might have missed significant arguments or findings from
literature in other academic disciplines or application areas. Nevertheless,
in the Discussion section, we have attempted to compare our findings
with those from other fields to enrich our understanding of various
perspectives on accountability, particularly regarding the aspect of
explanation.
Notes
1. e search was last conducted on July 30, 2023.
2. We c o nd u ct e d o u r s e ar c h a c ro ss d ie r en t r e se a rc h a r ea s t o g a th e r a s m u ch d at a a s
possible. Speci"cally, in the Web of Science, we examined the "elds of (1) public
administration, (2) information science and library science, (3) communication, (4)
international relations, (5) telecommunication, (6) social science and other topics,
(7) business economics, and (8) engineering. In EBSCO, we searched in (1) public
administration, (2) applied science & technology source, (3) library, information
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 29
science & technology, (4) social sciences, and (5) business. In SCOPUS, we looked
into (1) social science, (2) business management, (3) decision science, and (4) mul-
tidisciplinary "elds.
3. OECD (2003) defines the general government sector as “all units of central, state or
local government; all social security funds at each level of government; all non-mar-
ket non-profit institutions that are controlled and financed by government units.”.
Acknowledgments
An earlier version of this paper was presented at the Agile and Digital Governance Research
Colloquium in 2024. We extend our sincere gratitude to all participants at the colloquium for
their insightful and constructive feedback. We also deeply appreciate the anonymous reviewers,
whose invaluable comments significantly contributed to the refinement of this manuscript.
Disclosure statement
No potential conict of interest was reported by the author(s).
Notes on contributors
Qianli Yuan is a post-doctoral researcher in the School of International Relations and
Public Aairs at Fudan University. He received his Ph.D. in public management from the
Rockefeller College of Public Aairs and Policy, University at Albany, State University of
New York (SUNY-Albany). His areas of research are mainly related to digital governance,
public data, collaborative innovation, and co-production.
Tzu ha o C he n is a Ph.D. candidate at the Rockefeller College of Public Aairs and Policy
and serves as a Research Assistant at the Center for Technology in Government, both at
the University at Albany, State University of New York (SUNY-Albany). He will be joining
the Department of Public Policy and Administration at Florida International University
as an assistant professor in Fall 2025. His research interests include digital government,
arti"cial intelligence, public sector innovation, and smart cities.
References
Abusitta, A., Li, M. Q., & Fung, B. C. M. (2024). Survey on Explainable AI: Techniques,
challenges and open issues. Expert Systems with Applications, 255, 124710. https://doi.
org/10.1016/j.eswa.2024.124710
Ada Lovelace Institute, AI Now Institute, & Open Government Partnership. (2021).
Algorithmic accountability for the public sector (pp. 1–70). https://www.opengovpartnership.
org/documents/algorithmic-accountability-public-sector/
Aleksovska, M., Schillemans, T., & Grimmelikhuijsen, S. (2019). Lessons from "ve decades
of experimental and behavioral research on accountability: A systematic literature review.
Journal of Behavioral Public Administration, 2(2), 66. https://doi.org/10.30636/jbpa.22.66
Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R.,
Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023). Explainable Arti"cial
Intelligence (XAI): What we know and what is le to attain Trustworthy Arti"cial Intelligence.
Information Fusion , 99, 101805. https://doi.org/10.1016/j.inus.2023.101805
30 YUAN AND CHEN
Bannister, F., & Connolly, R. (2020). Administration by algorithm: A risk management
framework. Information Polity, 25(4), 471–490. https://doi.org/10.3233/IP-200249
Bell, A., Nov, O., & Stoyanovich, J. (2023). ink about the stakeholders "rst! Toward an
algorithmic transparency playbook for regulatory compliance. Data & Policy, 5, 8.
https://doi.org/10.1017/dap.2023.8
Bignami, F. (2022). Arti"cial intelligence accountability of public administration. e American
Journal of Comparative Law, 70(Supplement_1), I312–I346. https://doi.org/10.1093/ajcl/avac012
Bloch-Wehba, H. (2022). Algorithmic Governance from the Bottom Up. Brigham Young
University Law Review, 48(1), 69–136. Academic Search Complete.
Bovens, M. (2007). Analysing and assessing accountability: A conceptual framework.
European Law Journal, 13(4), 447–468. https://doi.org/10.1111/j.1468-0386.2007.00378.x
Bovens, M. (2014). Public accountability. In M. Bovens, R. E. Goodin, & T. Schillemans
(Eds.), e Oxford Handbook of Public Accountability. Oxford University Press. https://
doi.org/10.1093/oxfordhb/9780199641253.013.0012
Bracci, E. (2023). e loopholes of algorithmic public services: An “intelligent” account-
ability research agenda. Accounting, Auditing & Accountability Journal, 36(2), 739–763.
https://doi.org/10.1108/AAAJ-06-2022-5856
Brand, D. J. (2022). Responsible arti"cial intelligence in government: Development of a
legal framework for South Africa. JeDEM - eJournal of eDemocracy and Open Government,
14(1), 130–150. https://doi.org/10.29379/jedem.v14i1.678
Buhmann, A., & Fieseler, C. (2021). Towards a deliberative framework for responsible
innovation in arti"cial intelligence. Tec hn ol o g y i n S o c ie ty , 64, 101475. Scopus. https://
doi.org/10.1016/j.techsoc.2020.101475
Bullock, J. B. (2019). Arti"cial intelligence, discretion, and bureaucracy. e American
Review of Public Administration, 49(7), 751–761. https://doi.org/10.1177/0275074019856123
Bullock, J. B., Huang, H., & Kim, K.-C. (. (2022). Machine intelligence, bureaucracy, and
human control. Perspectives on Public Management and Governance, 5(2), 187–196.
https://doi.org/10.1093/ppmgov/gvac006
Bullock, J., Young, M. M., & Wang, Y.-F. (2020). Arti"cial intelligence, bureaucratic form,
and discretion in public service. Infor mation Polity, 25(4), 491–506. https://doi.
org/10.3233/IP-200223
Busuioc, M. (2021). Accountable arti"cial intelligence: Holding algorithms to account.
Public Administration Review, 81(5), 825–836. https://doi.org/10.1111/puar.13293
Carney, T. (2020). Arti"cial intelligence in welfare: Striking the vulnerability balance?
Monash University Law Review., 46(2), 23–51. Academic Search Complete.
Chen, T., & Gasco-Hernandez, M. (2024). Uncovering the results of AI Chatbot use in
the public sector: Evidence from US State Governments. Public Performance &
Management Review, 2024, 1–26. https://doi.org/10.1080/15309576.2024.2389864
Chen, T., Gascó-Hernandez, M., & Esteve, M. (2024). e adoption and implementation
of arti"cial intelligence Chatbots in public organizations: Evidence from U.S. State
Governments. e American Review of Public Administration, 54(3), 255–270. https://
doi.org/10.1177/02750740231200522
Chen, Y.-C., Ahn, M. J., & Wang, Y.-F. (2023). Arti"cial intelligence and public values:
Va lu e I m pa c ts a nd g ov er n a nc e i n t h e p u bl i c s e c to r. Sustainability, 15(6), 4796. https://
doi.org/10.3390/su15064796
Chung, I. H., Williams, D. W., & Do, M. R. (2022). For better or worse? Revenue fore-
casting with machine learning approaches. Public Performance & Management Review,
45(5), 1133–1154. https://doi.org/10.1080/15309576.2022.2073551
Cobbe, J., & Singh, J. (2020). Reviewable automated decision-making. Computer Law &
Security Review, 39, 105475. https://doi.org/10.1016/j.clsr.2020.105475
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 31
Considine, M., Mcgann, M., Ball, S., & Nguyen, P. (2022). Can robots understand welfare?
Exploring machine bureaucracies in welfare-to-work. Journal of Social Policy, 51(3),
519–534. https://doi.org/10.1017/S0047279422000174
Contini, F. (2020). Arti"cial intelligence and the transformation of humans, law and
technology interactions in judicial proceedings. Law, Technology and Humans, 2(1),
4–18. https://doi.org/10.5204/lthj.v2i1.1478
Crawford, K., & Schultz, J. (2019). AI systems as state actors. Columbia Law Review,
119(7), 1941–1972.
Criado, J. I., & Gil-Garcia, J. R. (2019). Creating public value through smart technologies and
strategies: From digital services to arti"cial intelligence and beyond. International Journal
of Public Sector Management, 32(5), 438–450. https://doi.org/10.1108/IJPSM-07-2019-0178
de Sousa, W. G., Melo, E. R. P. d., Bermejo, P. H. D. S., Farias, R. A. S., & Gomes, A.
O. (2019). How and where is arti"cial intelligence in the public sector going? A liter-
ature review and research agenda. Government Information Quarterly, 36(4), 101392.
https://doi.org/10.1016/j.giq.2019.07.004
Denyer, D., & Tran"eld, D. (2009). Producing a systematic review. In e Sage handbook
of organizational research methods (pp. 671–689). Sage Publications Ltd.
Donovan, J., Caplan, R., Matthews, J., & Hanson, L. (2018). Algorithmic accountability: A
primer. Data & Society.
Dubnick, M. J. (2014). Accountability as a cultural keyword. In M. Bovens, R. E. Goodin,
& T. Schillemans (Eds.), e Oxford handbook of public accountability. Oxford University
Press. https://doi.org/10.1093/oxfordhb/9780199641253.013.0017
Dunleavy, P., & Margetts, H. (2023). Data science, arti"cial intelligence and the third
wave of digital era governance. Public Policy and Administration, 2023, e09520767231198737.
https://doi.org/10.1177/09520767231198737
Dwivedi, R., Dave, D., Naik, H., Singhal, S., Omer, R., Patel, P., Qian, B., Wen, Z., Shah,
T. , M or g an , G ., & R an j an , R . ( 2023). Explainable AI (XAI): Core Ideas, Techniques,
and Solutions. ACM Computing Surveys, 55(9), 1–33. https://doi.org/10.1145/3561048
Engstrom, D. F., & Ho, D. E. (2020). Algorithmic accountability in the administrative
state. Yal e Jo u r na l o n R e g ul a t io n , 37(3), 800–854.
European Commission. (2018). Communication from the Commission to the European
Parliament, the European Council, the Council, the European Economic and Social
Committee and the Committee of the Regions—Coordinated Plan on Articial Intelligence.
European Commission.
Fernández-Loría, C., Provost, F., & Han, X. (2022). Explaining data-driven decisions made
by AI systems: e counterfactual approach. MIS Quarterly, 45(3), 1635–1660. https://
doi.org/10.25300/MISQ/2022/16749
Finer, H. (1941). Administrative responsibility in democratic government. Public
Administration Review, 1(4), 335–350. https://doi.org/10.2307/972907
Fink, K. (2018). Opening the government’s black boxes: Freedom of information and
algorithmic accountability. Information, Communication & Society, 21(10), 1453–1471.
https://doi.org/10.1080/1369118X.2017.1330418
Franzke, A. S., Muis, I., & Schäfer, M. T. (2021). Data Ethics Decision Aid (DEDA): A
dialogical framework for ethical inquiry of AI and data projects in the Netherlands. Ethics
and Information Technology, 23(3), 551–567. https://doi.org/10.1007/s10676-020-09577-5
Friedrich, C. J. (1940). Public policy and the nature of administrative responsibility. Public
Policy, 1(1), 3–24.
Giest, S., & Grimmelikhuijsen, S. (2020). Introduction to special issue algorithmic trans-
parency in government: Towards a multi-level perspective. Information Polity, 25(4),
409–417. https://doi.org/10.3233/IP-200010
32 YUAN AND CHEN
Grimmelikhuijsen, S. (2023). Explaining why the computer says no: Algorithmic trans-
parency aects the perceived trustworthiness of automated decision-making. Public
Administration Review, 83(2), 241–262. https://doi.org/10.1111/puar.13483
Hassija, V., Chamola, V., Mahapatra, A., Singal, A., Goel, D., Huang, K., Scardapane, S.,
Spinelli, I., Mahmud, M., & Hussain, A. (2024). Interpreting Black-Box Models: A
review on explainable arti"cial intelligence. Cognitive Computation, 16(1), 45–74. https://
doi.org/10.1007/s12559-023-10179-8
Henman, P. (2020). Improving public services using arti"cial intelligence: Possibilities,
pitfalls, governance. Asia Pacic Journal of Public Administration, 42(4), 209–221. https://
doi.org/10.1080/23276665.2020.1816188
Hiebl, M. R. W. (2023). Sample selection in systematic literature reviews of management research.
Organizational Research Methods, 26(2), 229–261. https://doi.org/10.1177/1094428120986851
Janssen, M., Hartog, M., Matheus, R., Yi Ding, A., & Kuk, G. (2022). Will algorithms
blind people? The effect of explainable AI and decision-makers’ experience on
AI-supported decision-making in government. Social Science Computer Review, 40(2),
478–493. https://doi.org/10.1177/0894439320980118
Jordan, M. I. (2019). Arti"cial intelligence – e revolution hasn’t happened yet. Harvard
Data Science Review, 1(1), 1–8. https://doi.org/10.1162/99608f92.f06c6e61
Jørgensen, A. M., & Nissen, M. A. (2022). Making sense of decision support systems:
Rationales, translations and potentials for critical reections on the reality of child
protection. Big Data & Society, 9(2), 163. https://doi.org/10.1177/20539517221125163
Jung, H., & Camarena, L. (2024). Street-level bureaucrats & AI interactions in public
organizations: An identity based framework. Public Performance & Management Review,
2024, 1–30. https://doi.org/10.1080/15309576.2024.2447352
König, P. D., & Wenzelburger, G. (2020). Opportunity for renewal or disruptive force?
How arti"cial intelligence alters democratic politics. Government Information Quarterly,
37(3), 101489. https://doi.org/10.1016/j.giq.2020.101489
Kuziemski, M., & Misuraca, G. (2020). AI governance in the public sector: ree tales
from the frontiers of automated dec ision -ma king in de moc ratic settings.
Te le co m mu ni c at io n s P ol i cy , 44(6), 101976. https://doi.org/10.1016/j.telpol.2020.101976
Li, B., Qi, P., Liu, B., Di, S., Liu, J., Pei, J., Yi, J., & Zhou, B. (2023). Trustworthy AI:
From principles to practices. ACM Computing Surveys, 55(9), 1–46. https://doi.
org/10.1145/3555803
Liberati, A., Altman, D. G., Tetzla, J., Mulrow, C., Gøtzsche, P. C., Ioannidis, J. P. A.,
Clarke, M., Devereaux, P. J., Kleijnen, J., & Moher, D. (2009). e PRISMA statement
for reporting systematic reviews and meta-analyses of studies that evaluate health care
interventions: Explanation and elaboration. PLoS Medicine, 6(7), e1000100. https://doi.
org/10.1371/journal.pmed.1000100
Liu, H.-W., Lin, C.-F., & Chen, Y.-J. (2019). Beyond state v loomis: Arti"cial intelligence,
government algorithmization and accountability. International Journal of Law and
Information Technology, 27(2), 122–141. https://doi.org/10.1093/ijlit/eaz001
Madan, R., & Ashok, M. (2023). AI adoption and diusion in public administration: A
systematic literature review and future research agenda. Government Information
Quarterly, 40(1), 101774. https://doi.org/10.1016/j.giq.2022.101774
Mansbridge, J. (2014). A contingency theory of accountability. In M. Bovens, R. E. Goodin,
& T. Schillemans (Eds.), e Oxford handbook of public accountability (pp. 55–68).
Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199641253.013.0019
McGregor, L. (2018). Accountability for governance choices in arti"cial intelligence:
Aerword to Eyal Benvenisti’s Foreword. European Journal of International Law, 29(4),
1079–1085. https://doi.org/10.1093/ejil/chy086
PUBLIC PERFORMANCE & MANAGEMENT REVIEW 33
Meijer, A., Lorenz, L., & Wessels, M. (2021). Algorithmization of bureaucratic organiza-
tions: Using a practice lens to study how context shapes predictive policing systems.
Public Administration Review, 81(5), 837–846. https://doi.org/10.1111/puar.13391
Mohseni, S., Zarei, N., & Ragan, E. D. (2021). A multidisciplinary survey and framework
for design and evaluation of explainable AI systems. ACM Transactions on Interactive
Intelligent Systems, 11(3-4), 1–45. https://doi.org/10.1145/3387166
Nam, J., & Bell, E. (2024). E&ciency or equity? How public values shape bureaucrats’ will-
ingness to use arti"cial intelligence to reduce administrative burdens. Public Performance
& Management Review, 2024, 1–34. https://doi.org/10.1080/15309576.2024.2419132
OECD. (2003). OECD Glossary of Statistical Terms – Government sector Denition. https://
stats.oecd.org/glossary/detail.asp?ID=1139
OECD. (2014). OECD Glossary of Statistical Terms – Public sector Denition. https://stats.
oecd.org/glossary/detail.asp?ID=2199
OECD. (2019). Articial intelligence in society. OECD. https://doi.org/10.1787/eedfee77-en
Parycek, P., Schmid, V., & Novak, A.-S. (2023). Arti"cial Intelligence (AI) and automation
in administrative procedures: Potentials, limitations, and framework conditions. Journal
of the Knowledge Economy, 15(2), 8390–8415. https://doi.org/10.1007/s13132-023-01433-3
Patil, S. V., Vieider, F., & Tetlock, P. E. (2014). Process versus outcome accountability. In M.
Bovens, R. E. Goodin, & T. Schillemans (Eds.), e Oxford Handbook of Public Accountability.
(pp. 69–89). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199641253.013.0002
Pencheva, I., Esteve, M., & Mikhaylov, S. J. (2020). Big Data and AI – A transformational
shi for government: So, what next for research? Public Policy and Administration,
35(1), 24–44. https://doi.org/10.1177/0952076718780537
Plant, J. F. (2011). Carl J. friedrich on responsibility and authority. Public Administration
Review, 71(3), 471–482. https://doi.org/10.1111/j.1540-6210.2011.02368.x
Plantinga, P. (2024). Digital discretion and public administration in Africa: Implications
for the use of arti"cial intelligence. Information Development, 40(2), 332–352. https://
doi.org/10.1177/02666669221117526
Rachovitsa, A., & Johann, N. (2022). e human rights implications of the use of AI in
the digital welfare state: Lessons Learned from the Dutch SyRI Case. Human Rights
Law Review, 22(2), 10. https://doi.org/10.1093/hrlr/ngac010
Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F., Breazeal, C., Crandall,
J. W., Christakis, N. A., Couzin, I. D., Jackson, M. O., Jennings, N. R., Kamar, E., Kloumann,
I. M., Larochelle, H., Lazer, D., McElreath, R., Mislove, A., Parkes, D. C., Pentland, A. ‘.,
… Wellman, M. (2019). Machine behaviour. Nature, 568(7753), 477–486. https://doi.
org/10.1038/s41586-019-1138-y
Ranerup, A., & Henriksen, H. Z. (2019). Value positions viewed through the lens of
automated decision-making: e case of social services. Government Information
Quarterly, 36(4), 101377. https://doi.org/10.1016/j.giq.2019.05.004
Redden, J. (2020). Predictive analytics and child welfare: Toward data justice. Canadian
Journal of Communication, 45(1), 101–111. https://doi.org/10.22230/cjc.2020v45n1a3479
Richardson, R., & Kak, A. (2022). Suspect development systems: Databasing marginality
and enforcing discipline. University of Michigan Journal of Law Reform, 55(55.4), 813–
883. https://doi.org/10.36646/mjlr.55.4.suspect
Robinson, S. C. (2020). Trust, transparency, and openness: How inclusion of cultural
values shapes Nordic national public policy strategies for arti"cial intelligence (AI).
Te ch no l og y in S oc i et y , 63, 101421. https://doi.org/10.1016/j.techsoc.2020.101421
Romzek, B. S., & Dubnick, M. J. (1987). Accountability in the Public Sector: Lessons from
the challenger tragedy. Public Administration Review, 47(3), 227. https://doi.org/10.2307/
975901
34 YUAN AND CHEN
Romzek, B. S., LeRoux, K., & Blackmar, J. M. (2012). A preliminary theory of informal
accountability among network organizational actors. Public Administration Review, 72(3),
442–453. https://doi.org/10.1111/j.1540-6210.2011.02547.x
Russell, S. J. (2019). Human comp atible : Articial intel ligence an d the pro blem of control . Viking.
Saeed, W., & Omlin, C. (2023). Explainable AI (XAI): A systematic meta-survey of cur-
rent challenges and future opportunities. Knowledge-Based Systems, 263, 110273. https://
doi.org/10.1016/j.knosys.2023.110273
Sætra, H. S. (2020). A shallow defence of a technocracy of arti"cial intelligence: Examining
the political harms of algorithmic governance in the domain of government. Technology
in Society, 62, 101283. https://doi.org/10.1016/j.techsoc.2020.101283
Saldaña, J. (2015). e coding manual for qualitative researchers. In e coding manual
for qualitative researchers.(3E). SAGE.
Selten, F., & Meijer, A. (2021). Managing algorithms for public value. International Journal of Public
Admini stration in the D igital Age , 8(1), 1–16. https://doi.org/10.4018/IJPADA.20210101.oa9
Su, S., & Wang, R. (2024). Shirking when bureaucratic accountability is due? Evidence
from managing severe safety accidents. Public Performance & Management Review,
2024, 1–24. https://doi.org/10.1080/15309576.2024.2335298
van den Homberg, M. J. C., Gevaert, C. M., & Georgiadou, Y. (2020). e changing face
of accountability in humanitarianism: Using arti"cial intelligence for anticipatory action.
Politics and Governance, 8(4), 456–467. https://doi.org/10.17645/pag.v8i4.3158
van Noordt, C., & Misuraca, G. (2022). Arti"cial intelligence for the public sector: Results
of landscaping the use of AI in government across the European Union. Government
Information Quarterly, 39(3), 101714. https://doi.org/10.1016/j.giq.2022.101714
Ve re i nt e N at i on en E d . ( 2022). United Nations E-Government Survey 2022: e future of
digital government. United Nations.
Wa ld m an , A . , & M a rt i n , K . ( 2022). Governing algorithmic decisions: e role of decision
importance and governance on perceived legitimacy of algorithmic decisions. Big Data
& Society, 9(1), 449. https://doi.org/10.1177/20539517221100449
Williams, R., Cloete, R., Cobbe, J., Cottrill, C., Edwards, P., Markovic, M., Naja, I., Ryan,
F. , S i n g h , J . , & P a n g , W. ( 2022). From transparency to accountability of intelligent systems:
Moving beyond aspirations. Data & Policy, 4(e7), 1–23. https://doi.org/10.1017/dap.2021.37
Wirtz, B. W., Langer, P. F., & Fenner, C. (2021). Arti"cial intelligence in the public sec-
tor – A research Agenda. International Jour nal of Public Administration, 44(13), 1103–
1128. https://doi.org/10.1080/01900692.2021.1947319
Wirtz, B. W., Weyerer, J. C., & Geyer, C. (2019). Arti"cial intelligence and the public
sector – Applications and challenges. International Journal of Public Administration,
42(7), 596–615. https://doi.org/10.1080/01900692.2018.1498103
Wirtz, B. W., Weyerer, J. C., & Sturm, B. J. (2020). e Dark Sides of arti"cial intelligence:
An Integrated AI governance framework for public administration. International Journal
of Public Administration, 43(9), 818–829. https://doi.org/10.1080/01900692.2020.1749851
Ya n g , K . , & D u b n i ck , M . ( 2016). Introduction: Accountability study moving to the next
level. Public Performance & Management Review, 40(2), 201–207. https://doi.org/10.10
80/15309576.2016.1266880
Yo u n g , M . M . , B u l l o c k , J . B . , & L e c y , J. D. ( 2019). Arti"cial discretion as a tool of gov-
ernance: A framework for understanding the impact of arti"cial intelligence on public
administration. Perspectives on Public Management and Governance, 2(4), 301–313.
https://doi.org/10.1093/ppmgov/gvz014
Zuiderwijk, A., Chen, Y.-C., & Salem, F. (2021). Implications of the use of arti"cial intel-
ligence in public governance: A systematic literature review and a research agenda.
Government Information Quarterly, 38(3), 101577. https://doi.org/10.1016/j.giq.2021.101577
