Government Information Quarterly 37 (2020) 101509

Contents lists available at ScienceDirect

Government Information Quarterly

journal homepage: www.elsevier.com/locate/govinf

Assessing the public policy-cycle framework in the age of artificial T

intelligence: From agenda-setting to policy evaluation
⁎
David Valle-Cruza, , J. Ignacio Criadob, Rodrigo Sandoval-Almazánc, Edgar A. Ruvalcaba-Gomezd
a
Department of Engineering, Autonomous University of the State of Mexico, Toluca, Mexico
b
Department of Political Science, Autonomous University of Madrid, Madrid, Spain
c
Department of Public Administration, Autonomous University of the State of Mexico, Toluca, Mexico
d
Department of Public Policy, University of Guadalajara, Guadalajara, Jalisco, Mexico

1. Introduction The challenge of AI in the public policy-cycle, regarding this highly

technological, changing and complex context, is to make public ad-
Nowadays, Artificial Intelligence (AI) has become a top priority on ministrations faster, more efficient, precise, transparent, and responsive
the agenda of different nations around the world. AI may lead to to the needs of citizens. However, we do not know what the implica-
transformative applications within a wide range of industrial, in- tions of AI in the policy process are. Given how recent the trend is,
tellectual and social applications, far beyond those generated by pre- many governments are waiting to understand AI applications and find
vious industrial revolutions (Dwivedi et al., 2019). Countries such as important applications to their field of research (de Sousa, de Melo,
China, the United States of America, and most of the European Union Bermejo, Farias, & Gomes, 2019). Based on the current literature on AI
have already implemented AI techniques to improve internal govern- in public policy, public administration, political science and depart-
ment processes, the provision of services and interaction with citizens, ments for international relations, a selection of practical cases are stu-
or have developed a national strategy for the implementation of AI died through the lens of the policy-making cycle (Bridgman & Davis,
(Craglia et al., 2018). AI has the potential to change different aspects of 2003; Howlett, McConnell, & Perl, 2017; Howlett, Ramesh, & Perl,
government, including processes, interaction with citizens, service de- 2009). The objective is to understand the implementation of AI and
livery, decision-making and public policy design and evaluation (Sun & work toward a dynamic framework. Therefore, the main objective of
Medaglia, 2019; Valle-Cruz, 2019). In fact, AI could help humans in this investigation is to assess the public policy-cycle framework in the
decision-making, in understanding and deriving meaningful results age of artificial intelligence (AI), regarding the actual and expected
from complex big data nexus (Höchtl, Parycek, & Schöllhammer, 2016). changes that these emerging technologies will bring about at different
However, selection, design, implementation, and use of AI in govern- moments in the policy-making process.
ment requires the collaboration of different specialists in the field, The structure of the article is as follows: The second section shows
implies massive automation of processes within government, and en- concepts, techniques and theory related to AI. The aim here is to further
tails involving new professionals to promote AI implementation. our understanding of the relationship between the public policy fra-
The scientific study of AI began in the 1940s, in the field of com- mework and AI techniques. The third section describes AI in the context
puter science. However, its application and consequences in the public of public policy, considering potential opportunities, challenges, and
sector have not been studied in depth. AI in government involves the negative implications. Furthermore, it analyses cases related to the
design, building, use, and evaluation of intelligent algorithms, robotics, public policy-cycle. The fourth section introduces the features of AI
and computational techniques to improve the management of public which seek to make the public policy-cycle (DPPC) more dynamic. The
agencies (Desouza, 2018). At present, few studies report that govern- fifth section creates a basis for discussion. Finally, the last section
ments and institutions are responding to the demands of a rapidly- summarizes the conclusion of the article.
changing society. Some facts that governments need to consider are a
product of the growing need and demand for public services, costs and 2. Artificial intelligence: Concepts and theory
time implementations, personalization, or quality of services for im-
proving public value: AI has the potential to improve governments. AI- This section aims to report on the current general state of Artificial
based technologies are promoting faster and more specific answers to Intelligence (AI). We shall especially focus on AI in public administra-
increasingly complex societal problems. tion and political science. First, we present the paper's research

⁎
Corresponding author.
E-mail addresses: [email protected] (D. Valle-Cruz), [email protected] (J.I. Criado), [email protected] (R. Sandoval-Almazán),
[email protected] (E.A. Ruvalcaba-Gomez).

https://doi.org/10.1016/j.giq.2020.101509
Received 15 October 2019; Received in revised form 26 May 2020; Accepted 9 July 2020
Available online 28 July 2020
0740-624X/ © 2020 Elsevier Inc. All rights reserved.
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

approach. Second, we tackle the notion of AI and the most important The creation of intelligent software (algorithms) and hardware
systems currently in operation. Then, we focus on the concept and (machines) based on AI uses different techniques. Some techniques are
techniques of AI in the framework of a literature review regarding useful for promoting learning, others evolve, and others are based on
implications from the perspective of public administration, political data analysis, and robotics (Floreano & Mattiussi, 2008; Jackson, 2019;
science, and social science journals. This section provides the con- Russell & Norvig, 2016; Tanimoto, 1987). All these AI techniques have
ceptual groundwork of the article. the potential to impact on organizations. Additionally, organizations
can use these emerging technologies for mass process automation, cost
2.1. Research approach and error reduction, increasing efficiency and competitiveness, creation
of social value, and fraud avoidance (Valle-Cruz, 2019). For example,
Due the incipient nature of the AI in Public Administration, we have intelligent agents have the potential to improve government-to-citizen
designed our research approach in two stages. The first stage is a the- interaction. Text mining, sentiment analysis, expert systems, and ma-
oretical and systematic exploration of documents and scholarly articles chine learning base their operation on techniques such as genetic al-
on specialized literature in the field of AI. In so doing, we further our gorithms, artificial neural networks, fuzzy logic, and statistics. Robotics
understanding of AI conceptualization and techniques. This first ap- alludes to hardware, however, there are robots programmed on the
proach helps us to classify and select information regarding AI, tech- Internet (software), which lack physical components. Techniques such
niques, and public policy and administration research, in order to make as artificial vision, autonomous vehicles, and virtual reality base their
connections between theory and practice. This first part of the research algorithms on software techniques embedded in hardware (Valle-Cruz
process generates the conceptual and theoretical groundwork for the et al., 2019). Nevertheless, any kind of robot bases its operation on the
following stage. characteristics of living beings, for example a mechanical arm.
In fact, research on AI in public sector organizations and govern- AI design and implementation in the public sector requires a
ments is not new. The first paper on the role of AI in public adminis- knowledge of different areas, and of new professions such as data sci-
tration dates back to 1989. This research espouses the idea that expert entists, AI engineers, and chief algorithm/AI officers. Among other
systems are essential for the efficiency of public agencies (Hadden, disciplines, an intelligent agent requires the use of evolutionary tech-
1989). In a seminal study, Barth and Arnold (1999) argued that the niques, artificial neural networks, and neuroscience. Thus, some areas
field of AI would lead to a new level of computation, that it would act as that are needed for AI design are: biology, neuroscience, linguistics,
an autonomous agent and learn independently, assess its environment programming, statistics, economics, operations research, game theory,
and “think” based on values, motives, and emotions. In addition, they among others: AI is a multidisciplinary area (Bredereck et al., 2014).
mentioned that AI would bring the greatest benefits to government Consequently, there is a variety of techniques and tools which are
decision-making, but would generate domination by elites who appro- useful for the implementation of AI, some of which are already in op-
priate such technology. eration in the public sector. The following section discusses the litera-
The second stage of the research approach is based on the analysis ture on AI techniques used in the public sector and administration re-
of selected cases. These cases focus on the different stages of the policy- search, and how this discipline may contribute to the diffusion of this
making cycle in order to further our understanding of the interplay new technological wave.
between AI and the public policy-cycle. The policy-cycle has become
the most widely applied framework for organizing and systematizing 2.3. Artificial intelligence in public policy and administration research
research on public policies. It focuses attention on generic character-
istics of the policy process rather than on specific actors or institutions Literature on AI in public policy and administration, political sci-
or particular substantial problems and respective programs. ence and international fields is scarce. Nonetheless, it needs to be ad-
This beyond-the-cycle approach defines the limits of the policy- dressed to understand this emergent area. At present, there is growing
cycle approach. A policy-cycle is a first approximation in the com- speculation concerning the dangers and benefits of AI, because there is
plexity of employers and, therefore, provides a guide for public action. very little empirical research in the public sector to corroborate such
No policy model can claim to be universally applicable since each speculation (Dwivedi et al., 2019; Sun & Medaglia, 2019; Wirtz &
policy process is based on different logical frameworks within govern- Müller, 2019). The following paragraphs provide a literature review
ment institutions (Bridgman & Davis, 2003). Accordingly, in this second evaluating emerging research on AI in politics and public administra-
stage, we analyzed practical cases with the available information at that tion. From here, we will focus on the potential of AI techniques and
time to provide empirical evidence for our proposal of a new Dynamic systems in the policy-making process.
Public Policy-Cycle (DPPC). Therefore, this study consists of a two-stage The last decade witnessed a substantial rise in interest in studying AI
process oriented toward defining our DPPC model, as well as toward in government. The first efforts were driven by the boom of machine
debating the traditional approach to the policy-cycle model. learning, AI initiatives around the world, and, as a consequence of re-
cent advances in computing power, the availability of data, and new
2.2. What is artificial intelligence? algorithms. This has led to major breakthroughs in the last decade.
Thus, the literature review conducted in this document is limited to
AI can be understood as computational intelligence (Poole, scientific articles published from 2010 to 2019 included on the Web of
Mackworth, & Goebel, 1998). Intelligent machines, therefore, have the Science database and scientific articles from the most cited journals in
capacity to learn, rationalize, and process instructions (Barth & Arnold, social sciences. To achieve this, first a logical search was carried out
1999). AI refers to the simulation of the behavior of living beings, based which included AI techniques identified in the conceptualization of AI.
on bio-inspired systems with models which simulate the behavior of These techniques have been used in the fields of public administration,
different kinds of animals or viruses (Floreano & Mattiussi, 2008). De- political science, public policy, and international relations. We applied
spite this fact, we lack a definitive concept of AI because it is constantly the logical search to the topic (including abstract, keywords, and in-
evolving alongside technological advances. In addition, in the field of dexed fields) and titles of the scientific articles. Finally, the logical
social science, there is a lack of knowledge regarding AI techniques, search used the following syntax:
since it is generally only related to robotics or machine learning. In later TS = ((“Artificial Intelligence” OR “Algorithms” OR “Artificial
stages of AI development, knowledge is acquired, in addition to beliefs, Vision” OR “Artificial Life” OR “Autonomous Vehicles” OR “Expert
desires, and intentions (Balke & Gilbert, 2014). AI is not only applied to Systems” OR “Classification Systems” OR “Robotics” OR “Artificial
machines (robots). AI also bases its operation on intelligent software Neural Networks” OR “Pattern Recognition” OR “Virtual Reality” OR
that encompasses a wide variety of computational techniques. “Data Mining” OR “Machine Learning” OR “Fuzzy Logic” OR “Natural

2
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

Language Processing” OR “Deep Learning” OR “Sentiment Analysis” OR

“Genetic Algorithms” OR “Intelligent agents”) AND (“Public Policy” OR
“Public Administration” OR “Political Science” OR “International
Relations”)) OR TI = ((“Artificial Intelligence” OR “Algorithms” OR
“Artificial Vision” OR “Artificial Life” OR “Autonomous Vehicles” OR
“Expert Systems” OR “Classification Systems” OR “Robotics” OR
“Artificial Neural Networks” OR “Pattern Recognition” OR “Virtual
Reality” OR “Data Mining” OR “Machine Learning” OR “Fuzzy Logic”
OR “Natural Language Processing” OR “Deep Learning” OR “Sentiment
Analysis” OR “Genetic Algorithms” OR “Intelligent agents”) AND
(“Public Policy” OR “Public Administration” OR “Political Science” OR
“International Relations”)).
Inclusion criteria focused on scientific articles contained in journals
on political science, public administration, public policy and interna-
tional relations. We excluded papers that were not related to AI tech- Fig. 1. Scientific Articles on Artificial Intelligence and Public Policy and
niques. After removing duplicates and articles that did not hold to the Administration, 2010–2019.
criteria listed above, this search led to an initial number of 29 refer-
ences. We added scientific articles from the most cited journals in social
that will drive innovation in government. In general, this research fo-
science. These journals traditionally publish in the field of electronic
cuses on studying challenges in the public sector and envisions AI as an
and digital government are included: Government Information Quarterly
emerging issue (Janssen & Kuk, 2016). Some applications of AI in
and Social Science Computer Review, where we found 20 papers related
public administration are related to the analysis of data through in-
to AI (Table 1).
telligent algorithms that will help social development, and enhance
Fig. 1 shows the number of scientific articles on AI techniques re-
decision-making, and forecasting (Meijer, 2017; Van der Voort,
lated to political science and international relations, as well as public
Klievink, Arnaboldi, & Meijer, 2019). In addition, AI systems will be
administration and policy fields from 2010 to 2019. There has been a
useful for the development of applications that improve citizen-oriented
notable increase in interest since 2016 as a result of the design of AI
services (Criado & Villodre, 2020). Nonetheless, there are some risks
strategies around the world, and the promising benefits of machine
related to algorithmic discrimination, opacity, human replacement, and
learning techniques for processing big data. Hence, an important fea-
the growth of the digital divide (Margetts, 2017; Margetts & Dorobantu,
ture of the scientific articles found is that most of them are based on
2019). Some computational techniques tackled in the literature review
policy documents or reports produced by NGOs, governments, inter-
are machine learning, intelligent agents, big data, robotics, autonomous
national organizations, and consulting firms. This implies that em-
vehicles, data mining, and chat-bots (Agarwal, 2018; Bredereck et al.,
pirical research and data from real implementation cases is still very
2014; Dobell & Zussman, 2018; Schuelke-Leech, Jordan, & Barry, 2019;
limited.
Sun & Medaglia, 2019; Vetrò, Santangelo, Beretta, & De Martin, 2019;
The existing literature on AI in public administration and policy as
Wirtz, Weyerer, & Geyer, 2018).
well as political science is chiefly focused on sophisticated technology
AI has the potential to modify different areas of government from
agriculture to public health (Castro & New, 2016). Wirtz et al. (2018)
Table 1 identify 10 AI applications: AI-Based Knowledge Management Soft-
AI literature review. ware, AI Process Automation Systems, AI Process Automation Systems,
Journal Scientific articles
Predictive Analytics & Data Visualization, Identity Analytics, Cognitive
2010–2019 Robotics & Autonomous Systems, Recommendation Systems, Intelligent
Digital Assistants, Speech Analytics, Cognitive Security Analytics &
Social Science Computer Review 15 Threat Intelligence. These technologies can impact different govern-
Government Information Quarterly 5
ment processes through automation. Additionally, data analysis leads to
American Journal of Political Science 2
Canadian Public Administration 2 valuable information which may be used by decision-makers.
Millennium-Journal of International Studies 2 There are some attempts to provide clarification on the different
Political Analysis 2 areas of potential impact of AI in the public sector. Wirtz et al. (2018)
All Azimuth-A Journal of Foreign Policy and Peace 1
classified AI in government into five categories: AI government service,
American Political Science Review 1
British Journal of Political Science 1
working and social environment influenced by AI, public order and law
International Affairs 1 related to AI, AI ethics and AI government policy. However, less is
International Studies Quarterly 1 known regarding design implications, implementation, and the use of
International Journal of Information Management 1 AI, especially in government. The use of AI in the public sector has been
Journal of Information Technology and Politics 1
driven by success stories reported by the private sector. Areas of public
Journal of Political Science Education 1
Local Government Studies 1 administration which have implemented AI in their processes include
Osterreichische Zeitschrift fur Politikwissenschaft 1 health, taxation, tourism and education (Agarwal, 2018;
Peace Economics, Peace Science and Public Policy 1 Androutsopoulou, Karacapilidis, Loukis, & Charalabidis, 2019; Castro &
Policy & Internet 1 New, 2016; Schuelke-Leech et al., 2019; Vetrò et al., 2019). Never-
Politics 1
Public Management Review 1
theless, scholarly discussion focused on public administration should
Research and Politics 1 move forward from terminological and doctrinal disputes to the study
Review of Policy Research 1 of anticipating the important consequences of AI for the future scholar
Revista del CLAD Reforma y Democracia 1 and professional practice in public policy (Dobell & Zussman, 2018).
Security Dialogue 1
Androutsopoulou et al. (2019) studied the use of chat-bots in the
Security Studies 1
Socio-Economic Review 1 Greek government's ministry of finance. Their findings indicate that AI
Transylvanian Review of Administrative Sciences 1 has the potential to improve communication between government and
Total 49 citizens through the use of techniques such as natural language pro-
cessing, machine learning and data mining. As can be seen, the
Source: Own elaboration.

3
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

government's immersion in AI will bring changes to different areas of artificial intelligence (AI) in government and public administration is
public policy. Additionally, some of the abovementioned challenges public policy in general, and the policy-cycle framework, in particular.
will need to be faced. AI research is frequently focused on the public This section is divided into two subsections. The first subsection de-
health system in certain countries. Watson for Oncology focus on the scribes the public policy-cycle in the age of AI; the second shows
personalized design of treatments for cancer patients, in this regard, practical cases related to AI techniques and public policy. Therefore,
Sun and Medaglia (2019) analyze the use of medical diagnostic appli- this section will open the door to the discussion of emerging changes in
cations through natural language, they provide four sets of guidelines the traditional approach to the policy-cycle and how they might be
for the governance of AI adoption in the public sector. transformed as a result of the intensive utilization of AI systems and
Other authors have analyzed different challenges and opportunities applications in different domains.
which have arisen from the impact of AI regarding different dimensions
of the public sector. Wirtz and Müller (2019) studied the use of AI in 3.1. The public policy-cycle in the age of artificial intelligence
public management structures in relation to its risks and side effects.
They argue that supported public management of intelligent machines Traditional approaches to public policy highlight that public inter-
requires essential ethical guidelines and policies for operation. They ventions go through a set of stages in a cycle. Each phase of the cycle
also mention some prospective benefits associated with public AI, such depends on the others, and changes in one phase affect the next cu-
as: improved information processing, accelerated case processing, im- mulatively. The policy-cycle, like any other, leads to a process of eva-
proved case assignment, and workforce substitution. Corvalán (2018) luation. A public policy, however, is not completed once its results have
has studied the transformation of public administration and the po- been analyzed. This analysis may lead to a new definition of the pro-
tential for generating smart governments that promote the effectiveness blem, causing the cycle to start over (Howlett et al., 2009). The policy-
of rights and inclusive technological development to ensure people's cycle has served as a basic reference that allows for systematization and
digital dignity. This is in spite of the idea that AI adoption could in- comparison of the various approaches and models in the field and for
crease the digital divide and inequality (Keynes, 2010). Mikhaylov, the assessment of the contribution to the discipline. At the same time,
Esteve, and Campion (2018) found an interplay between public sector the policy-cycle has been regularly criticized in terms of its theoretical
opportunities and cross-sector collaboration using artificial intelligence. construction, as well as in terms of its empiricism. However, it has
Savaget, Chiarini, and Evans (2019) describes the importance of poli- worked as a useful tool for the analysis of public policies (Jann &
tical participation and artificial intelligence in the public sector. Wegrich, 2007).
Although there are promising positive effects of AI on government, Our research proposes the framework of the policy-cycle to study
some scholars have studied the negative side of AI in the public sector. the implications of AI in the public sector. The public policy-cycle ap-
Among others, the negative implications for public administration are proach encompasses the different moments in which AI applications
discrimination, opacity based on algorithm complexity, privacy viola- can have an impact on the activity of public sector organizations.
tion, consolidation of socio-economic inequalities, or an increase in the Birckland (2016) considers the public policy-cycle as an appropriate
digital divide, AI dependence, the risks involved in public administra- framework for understanding the complexity of public decision-making
tion, and decision-making related to algorithmic bias in the design of AI processes, as well as the actors involved in such processes. Thus, ac-
for the public sector. In this sense, Vetrò et al. (2019) studied the tivities, actors and drivers of public policy can be considered within this
governance and ethics of AI, demonstrating the potential risks of dis- framework. Therefore, the utilization of AI in public policy-cycle
crimination in the use of data in rational AI agents and analyzed some comprises its different interrelated stages (Dunn, 2012). These stages
general ethical principles derived from the AI White Paper at the service are: agenda-setting, policy formulation, and decision making, policy
of the citizen of the Italian Government agency. Other negative effects implementation, and policy evaluation. In each of these stages, the use
associated with public AI include technology obedience and loss of of AI can be proposed in order to facilitate empirical observation. The
control, AI dominion and AI legitimacy, AI paternalism and AI decision- policy-cycle framework may facilitate the discussion of the implications
making, problems with cybersecurity and, privacy violation (Wirtz & of AI in different types of public organizations (local/municipal, re-
Müller, 2019). gional/state or national/federal), as well as in different areas of activity
Different challenges for public policy in the age of AI are related to (health, education, social benefits, security, tax, migration, etc.). These
legal and moral issues, ethical principles, legitimacy, transparency, activities work toward a more dynamic framework.
workforce substitution through intelligent automation, predictions, and Table 2 summarizes the public policy-cycle stages, the main issues
decision making. Also, public policies should include the analysis of to be considered in each of them, the challenges and opportunities
hardware and AI algorithms, as well as the generation of standards, derived from the application of AI, and the negative implications they
norms, and rules to avoid bias, unexpected results or bad decision- may entail in the public sphere. Below, we present the different stages
making (Schuelke-Leech et al., 2019). Other potential risks for smart of the policy-cycle process from the perspective of its interaction with
governance based on algorithms include the fact that it is open to the AI. Table 2 is based on the previous work of some scholars who pre-
influence of opacity. For this reason, intelligent algorithms must be viously studied big data applications with reference to the policy-cycle
transparent, based on giving citizens access to information about the approach. Giest (2017) and Höchtl et al. (2016) offer different results
actual content of the intelligent algorithms and how they work (Cerrillo from one another. Also, we consider other holistic perspectives
i Cerrillo i Martínez, 2019). Thierer, Castillo, and Russell (2017) have (Pencheva, 2018; Thierer et al., 2017; Wirtz et al., 2018) to compre-
investigated this problem, arguing that AI suddenly appeared in public hensively tackle AI in the public policy-cycle.
administration. Agenda setting and AI. One key aspect of agenda setting is how
Accordingly, initial research suggests that the opportunities for some issues, and not others, occupy the attention of the public and
public policy based on AI techniques will improve accountability, col- policy makers. Here, AI has a clear impact on and plays a key role in the
laboration, interoperability, efficiency, effectiveness, information pro- agenda-setting process. This process is performed by AI and frames
cessing, G2C, trust in decision making, among others. The following social problems and makes them suitable for the attention of policy-
section presents the public policy-cycle and its connection to AI. Our makers. Lazer (2016) suggests the risks of using social media as a means
analysis is based on practical case studies. of creating and following debates in democratic systems. Nonetheless,
digital media and online politics are becoming major sources of in-
3. Artificial intelligence and public policy formation and part of an agenda-setting generation. Then, using AI
algorithms, governments may track, almost instantaneously, emerging
One of the areas with most interest for future interaction with topics and promote agenda points to collect data from different social-

4
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

media sources. As Pencheva (2018) suggest, agenda-setting can be

Some governments may be excluded because of the diversity

Given the diversity of needs and the increasing digital divide,

The cultural and social factors, as well as the digital divide

improved in terms of accuracy, efficiency, and speed. Also, AI may

displacement of people from their workplaces and a high

The result can be the dehumanization of daily activities,
increase the legitimacy of agenda setting by giving governments the

and lack of connectivity in some countries generate

opportunity to foster more inclusive and diverse dialogue with citizens.

complexity in order to homogenize legislation.

However, governments with less infrastructure and technological cap-
abilities may be excluded from public policies based on AI. This would

dependence on intelligent technologies.

eventually lead to an increase in the digital divide. Another negative
of needs; not just technological ones.

the complexity of analysis increases.

implication could be privacy violations in the use of data to identify and
select social problems.
Bad decision-making or bias.

Policy formulation, decision-making and AI. The second stage in the

policy-cycle refers to the moment in which the policy agenda has been

Algorithm complexity.
Negative implications

defined and now political actors deal with the different policy options
Privacy violation.

Privacy violation. available. Here, AI may also have a noticeable impact as the debate on

AI dependence.
Discrimination.

different policy options moves forward and data and information on the
issue come to the fore. As an example, cancer diagnosis using image
Opacity.

recognition based on AI may help to decide how to invest in future

decisions that need to be taken regarding hospitals or even if new
specialist doctors are needed in a specific area of health (Sun &
lack of theoretical frameworks in data
Data obsolescence; data homogeneity;

Medaglia, 2019). At the same time, unstructured data may be available

Noise, digital divide, cumbersome

from different social-media tools (blogs, social platforms, microblog-

nature of the democratic process.

Improve information processing.

Legal and moral responsibilities.

Legal and moral responsibilities.

ging, etc.) and public organizations may retrieve ideas or feedback from
Noise and the digital divide.

these sources. Eventually, this process needs to be automated and AI

Goal displacement; loss of

Workforce substitution.

People's digital dignity.

will be an important tool for data mining, text analytics or social data
Intelligent automation.
Transparent software.

scouting. In this process, AI may help to reinforce accountability as

Ethical principles.

Ethical principles.

governments need to be more responsible for the decisions taken and

responsibility.

the policy options selected. Citizens and other political actors will be
Challenges

analytics.

watching in real time with data and information flowing to decide on

opportunistic political behaviors. In sum, AI can find patterns in data,
and this situation may lead to the discovery of new solutions that
support public policy, but also generate more complexity in decision-
Cost saving; productivity gains; reduced

monitoring; experimentation with new

making as a result of the discovery of new agents or causes for pre-

existing problems. Furthermore, there is the possibility of bad decision-
Improve policy analysis; real time
Improve information processing.

Improve information processing.

Accuracy, efficiency and speed;

making or bias that may lead to discrimination and lack of social trust
fraud; better service provision.
legitimacy and collaboration.

based on the opacity of algorithms.

Improve decision making.

Improve interoperability.
Accountability and trust.

Policy implementation and AI. This stage refers to the execution of

People's digital dignity.

actions to enforce the decisions taken in previous stages. This part of

the policy-cycle has been more accurately studied in the literature as it
service models.
Improve G2C.

Improve G2C.
Opportunities

Effectiveness.

includes the majority of projects developed and is up to date in terms of

Source: Own elaboration based on Pencheva, 2018; Thierer et al., 2017; Wirtz et al., 2018.

AI in public sector organizations. Utilization of AI and big data in the

implementation of a policy may be related to two aspects, following
Höchtl et al. (2016). On the one hand, they may facilitate the high-
lighting of issues that could be implemented with different levels of
How to deal with assessment of real-time data? How
make possibly unconventional problems arise in the

How to identify and select policy alternatives? How

How to identify and select social problems? How to

intensity. On the other hand, data and information sources for the
implementation of policies may be reinforced and sped up with AI
analytics. In this regard, continuous improvements in the process of
How to manage real time evidence from
implementation? How to deal with the

real-time data analytics may facilitate large-scale improvements of city

traffic as policy makers can benefit from automation. Consequently,
to promote alternative solutions?

policy implementation may improve service delivery regarding the in-

to perform holistic analytics?

dividualization and capacity to profile users according to socio-demo-

immediateness of change?

graphic features or behavioral patterns. Some tax agencies worldwide

agenda-setting process?

are profiling individuals and/or enterprises to further the fight against

Issues and questions

fraud by designing incentives oriented toward heightening tax-payers'

commitment. Also, police departments are using AI to monitor crime
and identify patterns of incidents of crime. At this point, data proces-
sing through AI can generate patterns that are useful for making pre-
Public policy-cycle in the age of AI.

dictions that could lead to the exclusion of vulnerable groups or gen-

erate greater control in society. Societies with greater digital divide
Policy formulation and decision

may have problems in the standardization and legislation of AI. At this

level, algorithm complexity could lead to the implementation of a bad
Policy implementation

policy.
Policy evaluation and AI. Evaluation is another phase in the policy-
Policy evaluation
Agenda-setting

cycle with a high potential to be transformed by AI in public admin-

making

istration. In this case, evaluation may be altered by the capacity to

detect irregularities and produce caveats when required. Some authors
Table 2

suggest that evaluation using data and AI will transform the notion of
evaluation itself (Höchtl et al., 2016). In other words, AI and big data

5
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

will facilitate the embedding of evaluation in each phase of the policy-

Grothaus (2019), Sun and Medaglia

Algorithmic opacity, illegitimacy of

Identify citizens wherever they go.
Real-time continuous monitoring.

Correction and redesign of public

cycle, superseding the notion of evaluation only happening at the end of

Fast and intelligent evaluation.

Pattern and facial recognition.

the policy process. Therefore, AI will make evaluations that are deliv-

Social control based on data.

Interpreting, interacting and

Flight status and find gates.

ered in real time from the very beginning of the implementation, giving
birth to the notion of continuous evaluation in governments and public

Knowledge creation.
administrations. Nonetheless, this AI and data-driven approach to

Policy evaluation

decision making.
evaluation may lead to a lack of substantive theory for understanding

Engagement.
the implications of data (something that is already happening in ap-

evaluating

Big data
plied-social-science research). To put it differently, if governments and

(2019)

policy.
public administrations address AI and big data, we may be looking too
closely at very specific and immediate situations and problems, instead
of holistic situations and underlying causes of problems. And this si-

Emergency responders can be very rapidly updated on an

Big data, robots, algorithms, machine learning and deep

Predictions made about which areas of a town are most

tuation may be exacerbated in cases in which accuracy and homo-

Baraniuk (2015), Soumik (2018), Pesapane, Volonté,

Personalization, interpretation, and identification of

geneity of data is not guaranteed. At the same, AI in the evaluation
stage, may cause some major problems that public policy can face.

likely to be worst affected by an earthquake.

These include dehumanization, complete dependence on AI, algo-

Better Government to Citizen interaction.

rithmic biases that influence future decision making, and segregation of
the poor and people with limited access to the most advanced tech-

Easy human-computer interaction.

nologies.

Predictive maintenance systems.

Codari, and Sardanelli (2018)
Consequently, AI is changing the policy-cycle process and, in one

Predicting and interpreting

Violation of privacy rights.

way or another, opening up new opportunities, but also challenges and

Policy implementation
negative implications, from agenda setting to evaluation processes.

Fast implementation.

emerging situation
Meanwhile, AI and the capacity to learn from public policy are emer-

particular cases.
ging as sources of new opportunity in the public sector. Data is at the
heart of public sector organizations and AI-mediated activities are

Save lives.
learning.
based on learning processes nurtured by data-intensive activities. The
new opportunities for AI in the public sector derive from the data-
driven nature of public institutions. At the same time, the challenges
also come from the same source. In fact, looking at the public sector

Use of dynamic models for learning and adaptation of policy

Robotics, Internet of Things, Drones, and Big Data Analytics.

Good (2016), Joyner-Roberson (2019), Hwa (2018), Cath,
(and the policy-cycle) holistically, it seems that the public sector

Forecasting and designing of models to simulate future

landscape is rapidly changing. This includes service models and human
behavior inside and outside public organizations. The following para-
Wachter, Mittelstadt, Taddeo, and Floridi (2018)

graphs support this conceptual approach with empirical evidence from

Keeping emergency responders out of danger.

Prediction and recommendation algorithms.
real public sector settings.
Policy formulation and decision making

Assistance for emergency responders.

3.2. Cases of artificial intelligence in the public policy-cycle

Predicting required resources.

Responding to cyber threats.

The cases were selected by carrying out a documentary analysis of e-
Discovering and predicting

government journals for practitioners and by looking for practical

Cyber-attack prevention.
guides related to AI in government that address public policy issues,
published between 2014 and 2019. At present, some of them are at an Help in emergencies.
early stage of development and some others are on-going cases at dif-
formulation.

ferent stages of the policy-cycle process. Then we analyzed the cited

situations.

sources in these documents to discover the implications for the public

policy-cycle. For case selection, we established some criteria: a) cases
must be clearly involved in a specific area/domain of the public sector,
Strengthen markets, boost economic growth
Track, report and map information related

and b) cases must show evidence of the development of public policy

based on AI. See Table 3 for a brief description of the implications of
Insight extraction from large datasets
Abnormalities and pattern detection.

each case at the different stages of the policy process.

Greenemeier (2014), Hwa (2018),

Monitor diseases based on citizen

McKelvey and MacDonald (2019)

Discover new solutions through

Big Data and Machine Learning

Analyze large amounts of data.

After analyzing these practical cases, we found different evidence

Monitoring and discovering

and conclusions that are categorized within the policy-cycle frame-

and global competition.
More efficient agencies.

work. The main convergence observed is the dynamism in the devel-

to public health crises,

opment of public policies boosted by AI. The main discrepancy ob-

served concerns the regulatory and data-protection perspective in the
Agenda setting

collaboration.

development of AI policies. Then, we developed the characteristics of

Cases of AI in the public policy-cycle.

simulations.
Real-time

the cases analyzed by following the four stages of the public policy-
cycle.
Agenda-setting cases. AI is changing the way agenda setting oper-
Source: Own elaboration.

ates nowadays. The selection, classification and prioritization of public

Typology of applications

problems is performed by machine learning. Algorithms search on large

Application features

databases. In the near future, AI will highlight important problems,

unfold hidden needs or problematic issues, and identify key elements
Consequences
Techniques

that could affect on-going public policies. Also, the agenda-setting

phase will be influenced by on-going citizen participation, will use
Table 3

Cases

crowdsourcing and big data technologies to track and report new pos-
sible outcomes from society that could be topics for the agenda.

6
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

Our cases point in that direction. For example, the Health Map to technological devices to prevent cyber-attacks (Cath et al., 2018). In
track the Ebola Outbreak relies on disease surveillance websites that act this manner, AI goes beyond just analyzing data, finding patterns or
as key hubs for information processing. These sites use a combination of new solutions, because it acts as an intelligent technological artifact
AI software and human expertise to track, report and map information that assists in decision-making and in the formulation of public policies,
related to public health crises, often faster than government ministries taking a step toward prediction.
and international organizations can respond (Greenemeier, 2014). Policy implementation cases. AI will improve policy implementa-
A similar case of improved agenda setting is the National Police Lab tion in several ways, including the speed of processing. However, this
AI in the Netherlands. The purpose of this case is to develop state-of- step requires budgets, progress tracking, and feedback. Most routines
the-art AI techniques to improve safety in the Netherlands by working can be automated by expert systems, robotics, autonomous vehicles,
on techniques across the full breadth of AI. Practitioners and scholars of and extended reality. The latter, especially, can produce some simula-
the University of Amsterdam use machine-learning techniques to ex- tions of policy implementations and help decision makers to refine data,
tract the right information from different sources of Big Data. Another and process decisions before implementation.
ally is Utrecht University, which will focus on models from symbolic AI Some cases are leading in this direction. In San Mateo County,
that allow for the reasoning and communication of information. California, which lies on the San Andreas fault, an algorithm, con-
Another example of agenda setting, also in the Netherlands, is using AI taining public data such as age, type and construction materials of in-
algorithms to analyze huge amounts of investigation requests. In this dividual buildings, allows seismic data and structural knowledge of
way, police officers can sieve out important intelligence on potential buildings to be used to judge which parts of a city will be most at risk
threats (Hwa, 2018). and prioritize rescue efforts. The algorithm effectively predicts how
Finally, the case of Artificial Intelligence Policy Innovations at the buildings will react to shock waves and maps damage. Emergency re-
Canadian Federal Government is a dynamic opening process which aims sponse teams can analyze this information to identify where the largest
to establish the need to insert the vision that algorithms can strengthen number of people affected is (Baraniuk, 2015).
markets, boost economic growth and global competition (McKelvey & Another relevant insight to the implementation phase of the policy-
MacDonald, 2019). These kinds of technologies allow for the mon- cycle is found in the case of Dubai Police Force, which has built smart
itoring of big data, discovering patterns and new solutions. These are police stations run entirely by AI-powered robots. The smart police
useful for improving and quickening agenda setting. Thus, this kind of station, which is in operation 24/7, provides services like crime re-
technologies has the potential to reduce costs and errors at an early porting, lost-and-found requests, and victim support. Additionally, it
stage of the policy-cycle process. receives work and income complaints and home security requests.
Policy formulation and decision-making cases. The new dynamic of On the other hand, India Ray has developed a new predictive
policy formulation and decision-making with AI will increase the speed maintenance system to help its railways prevent accidents by harnes-
of this step. Normally, this stage of the policy-cycle could take several sing artificial intelligence (AI). The device will be equipped with a high-
weeks or months, depending on the level of government or the com- definition camera to scrutinize parts of train coaches in real time and
plexity of the problem. However, with the use of AI, this process could transmit insights to relevant teams over a Wi-Fi connection. The robot
be reduced from days to hours. Some techniques such as artificial named USTAAD (acronym for Undergear Surveillance Through Artificial
neural networks, and evolutionary computation (genetic algorithms, Intelligence Assisted Droid) is changing Indian Railways in terms of
evolutionary strategies, and genetic programming) could help with the safety, growth, and efficiency. AI has to be harnessed to find digital
formulation of policies and decision making. In the near future, a innovations for a better customer interface and better service delivery
government advisor – robo-advisor – will forecast several decision op- (Soumik, 2018).
tions, their public value for society and the possible budget for each Finally, the case of the Food and Drug Administration (FDA) in the
decision. All these innovations might help the decision-maker to choose U.S.A. has started using machine learning and deep learning to identify
the best option for their own political purposes or according to their violations of privacy rights and cyber-attacks (Pesapane et al., 2018). At
limitations and context. this stage, AI generates predictions that assist with the implementation
In our analysis, we identified some cases that have used this kind of of public policies, promoting innovation and efficiency, and improving
artificial technology to improve decision making. The case of AUDREY, government to citizen interaction and interoperability between agen-
the Assistant for Understanding Data through Reasoning, Extraction, and cies.
Synthesis is part of an initiative to innovate new ways to keep fire- Policy evaluation cases. The final step of the policy-cycle will also
fighters, police, paramedics, and other rapid response teams safe be affected by AI. At this stage of the cycle, AI could help by providing
through increased awareness of their environment and communication real time feedback, instant solutions and simulations to improve im-
skills. AUDREY uses AI techniques that can track an entire fire team, plementation. This is achieved by using a mixture of techniques such as
sending relevant signals to people and helping to make recommenda- artificial neural networks, fuzzy logic, and intelligent systems to gen-
tions on how they might work together (Good, 2016). erate alternatives. Additionally, data mining, deep learning, and
A second example is the Indiana law which specifies that police learning classifiers are used to provide instant feedback on the process
departments can use drones for search-and-rescue efforts, to record coming from sources like citizens' satisfaction and by monitoring the
crash scenes, and to help in emergencies. Otherwise, a warrant is re- progress of policy implementation.
quired to use a drone (Joyner-Roberson, 2019). A similar effort can be Several cases point in this direction. For example, in China, face-
found in Australia. This country is in the process of building a national recognition kiosks have been implemented at airports to accurately
network of cyber-threat-sharing centers to detect and protect its citizens identify flight plans, as well as providing a map and instructions for
from malicious cyber-attacks. In these sharing centers, cyber security getting to the gate. The kiosk identifies the person in seconds and any
experts from academia, businesses and government can pool sensitive person could be identified by way of the social credit system and gov-
data on cyber threats in an online sharing portal. They then use big data ernment decisions (Grothaus, 2019). This is helpful for the evaluation
analytics to detect patterns of malicious cyber activity and implement of public policy in the fields of security and migration.
adaptive, data-driven ways of responding to cyber threats (Hwa, 2018). Another example is the case of the Innovation Center for Artificial
During the formulation of public policy, the European Agency for Intelligence (ICAI). This is a Dutch national network focused on joint
Robotics and AI expresses a collaborative vision of different social technological development which involves academia, industry and
sectors. There is concern regarding equity biases and software-regula- government in the area of AI. It gathers collective work from other
tion issues. This case is representative of a growing interest in the national labs like National Police Lab AI: Dutch National Police, AIRLab
protection of personal data (sensitive information) and the use of Delft (AI for Retail Robotics); Thira Lab (Imaging), and AFL (AI for

7
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

Fintech: intelligent algorithms to assist in finance). Accordingly, AI represents a technological disruption because it
Finally, the case of IBM Watson Health for public healthcare in China transforms the policy-process as we knew it in the past. It has the po-
highlights the idea of change in public employment. The problem of tential to analyze big data, find patterns and information that humans
opaque algorithms is that they delegitimize decision making. cannot. It may perform routine tasks more accurately and without
Additionally, there are ethical conflicts. Monitoring is perceived as an breaks, giving humans the freedom to do creative or innovative work
important idea in evaluation, which usually shows the social gap in and to generate strategies in decision making. Also, AI could allow
innovation processes (Sun & Medaglia, 2019). At this level, AI facil- humans to take advantage of its benefits to improve their capabilities.
itates and accelerates the evaluation of implemented public policies This might lead to a complex framework for social inclusion in the
based on data analysis, simulations, and intelligent algorithms. On the policy-cycle. However, through methodologies that simplify complexity
one hand, they lead to welfare and efficiency, but on the other hand, and democratize access to co-production mechanisms, it is possible to
they are invasive and lead to social control. boost the policy-cycle. From the literature review and case analysis we
described in the previous sections, we observe features of this dyna-
4. Toward a dynamic public policy-cycle framework mism, mainly in the capacity for convergence and the evaluation of
policies at the different stages of the cycle.
Our previous section showed practical cases of government orga- Broadly speaking, we identify some dimensions that are different in
nizations around the world that are adapting their public policies by the new DPPC model. The policy-cycle in the age of AI follows a dif-
using Artificial Intelligence (AI) techniques at the different stages of the ferent flow of steps. Rather than the previous conception of a single
policy-cycle. New found dynamism in the public policy-cycle as a result cycle, which requires several iterations to produce a public policy, the
of AI systems and applications may allow for the generation of better, impact of AI will evolve in an incremental and spiral fashion. It will
more inclusive and more efficient solutions to complex societal pro- give feedback at each stage of the cycle, analyzing data with AI tech-
blems. As our cases denote, AI has the potential to automate repetitive, niques and simulations. There will be no need to wait until the im-
low-value added work done by humans. This helps people to work more plementation phase to have results for evaluation. Now, AI can bring
efficiently. Repetitive and tedious work can be automatized, allowing about scenarios for each stage of the public policy-cycle in almost real
people to do new creative things, and improving human capacities time. Based on our cases, we found that the speed of the process and the
(Mergel, Gong, & Bertot, 2018). From our case analysis, this section integration of the elements of the policy-cycle turns the iteration stage
presents an emerging approach named “Dynamic Public Policy-Cycle” into a more dynamic process.
(DPPC). This refers to the on-going changes at the different stages of the Instead of the linear cycle, the spiral form represents an incremental
policy-making process based on AI. The DPPC derives from the gov- perspective of each step of the policy-cycle. Our exploratory case stu-
ernment's desire for efficiency and involves new organizational struc- dies showed that the speed produced by AI techniques changes each
tures in the form of digital service teams that can respond faster to the step, rendering each step almost instant. For example, agenda setting
ad hoc needs of their internal government clients, transform organiza- could occur at the same time as policy evaluation takes place; policy
tional culture and methods of collaboration, as well as quickly adapt to formulation and decision-making could happen almost at the same time
changing need and demands (Mergel, 2017; Soe & Drechsler, 2018). as policy implementation. Just “one click”, in the intelligent algorithms,
The following paragraphs will expand the rationale of this approach could divide each process. This spiral loop could be included in a deep-
based on the transformations derived from the utilization of AI systems learning algorithm, with neural network techniques, genetic algorithms
and techniques. or some others to ensure incremental intelligence, efficiency and speed
Agile government is based on software engineering methods like (see Fig. 2).
scrum, adapting changes on-the-fly, and working in cellular and net- Besides, each stage of the public policy-cycle can be evaluated
worked organizations, rather than in vertical, bureaucratic, and struc- faster, more accurately, and at a lower cost due to the massive data
tured environments. This methodology is oriented toward accelerating processing and simulations that artificial intelligence facilitates. Data
activities and processes, making them more dynamic in different or- processing techniques, through AI techniques, generate valuable in-
ganizational settings. The implementation of AI has the potential to formation for decision makers. As a result, decision makers gain useful
make the DPPC innovative, predictive, accurate, faster, flexible, cor- knowledge for the formulation of new strategies and re-formulation of
rectable, and customizable. Just as e-governance and e-government are societal problems. In other words, collecting data from the evaluation of
the continuation and evolution of traditional governance and govern- policy is to be expected and required in order to be able to agenda set,
ment, e-policy is the evolution of traditional policy-making (Höchtl then formulate several policy options and help the decision maker to
et al., 2016). However, with the influence of AI, we suggest the next make the best decision. This is followed by policy implementation and
step is a dynamic framework of e-policy: DPPC. working back from policy evaluation to create the next policy iteration.
The characteristics and ideas that converge in the formulation of Therefore, AI-input in the public policy frame has the potential to
policies are changing with the insertion of new technologies, which are come from data in the agenda-setting step. This data is transformed into
increasingly sophisticated and algorithm based. The speed of making knowledge for policy formulation and policy implementation. It be-
decisions in the policy-cycle is a variable that substantially affects the comes public value in terms of AI policy evaluation. Data should come
process. Also, the development of capabilities to evaluate results, as from citizens' feedback and/or other government sources. This might
well as to release the information boost the development of public lead to a more dynamic, creative and efficient public policy-cycle, re-
policy. Hence, dynamism in the policy-cycle refers to the acceleration of ducing error in policy selection and public spending. This intelligence
innovation processes and activities that are transforming the way in automation, innovation and augmentation is useful for the generation
which public policies are adopted, implemented, and evaluated. of AI-based public policy strategies in a world with more data avail-
Dynamism in the policy-cycle is aimed at filling the information gap ability and social, political, and economic uncertainties.
between the government and the social sectors and actors. Currently, For policy makers and public officials, the use of AI represents a
different societal actors have greater access to information due to the strategic option for the improvement of the policy-making process.
widespread use of emerging technologies, which has allowed for more Now, it is not only possible to consult people and to involve them in
symmetry in information and capacities between government and so- government action, but it is also possible to consult robots and in-
ciety (Janssen & Helbig, 2018). This balance of forces and the demo- telligent information systems to undertake a policy that affects the
cratization of information also impacts the policy-cycle, making it more public sphere. This encompasses emerging challenges for public man-
agile and dynamic, allowing the participation of actors such as experts agers and policy makers that have the potential to change the tradi-
and civil society organizations. tional approach to the public policy-cycle (Table 4).

8
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

more accurate, and more efficient during the public policy process. The
DPPC will facilitate the transitions between the stages of public policy,
dynamizing what Howlett et al. (2009) and Howlett et al. (2017) sus-
tain are the critical features of the classic policy-cycle. The result is the
redefining of social problems easily, with policy readjustments, either
through AI-simulations or AI-experiences concerning the agenda. Also,
the DPPC will be enhanced by the processing capacity of emerging
computer systems, such as quantum computing, and the new algorithms
that will be implemented for their analysis (Newcombe, 2019). This
situation is currently limited.
Also, the DPPC model can be assessed in terms of the different stages
of the policy process. In the agenda-setting stage, the detection of
problems could be preventive, not merely corrective. With this ap-
proach, it would be possible to make progress toward the solution of
more complex social problems. Besides, sophisticated technologies
based on AI allow access to more groups of actors interested in public
policies. Sophisticated AI-technologies also allow for the collection of
information from various sources, in real-time. This is useful in the
design of the public agenda. Logics within government institutions are
modified (Bridgman & Davis, 2003) in accordance with an approach
based on social welfare and the generation of public value.
Also, the policy formulation and decision-making stage could be
improved by the adoption of the DPPC model. Particularly, this stage
would be enhanced by many alternatives that exist through the massive
analysis of data, allowing the implementation of preventive and effec-
tive actions and the increase of public value (Criado & Gil-Garcia,
2019). Due to the use of algorithms based on AI, decision-making could
be more objective and decision-making less emotional or based on
decision-making management traits or behaviors (Schuelke-Leech et al.,
2019). This situation could lead to benefits for society, but also greater
inflexibility. Hence, following Weber's classical metaphor of the iron-
cage, AI might bring about a surveillance society (Zuboff, 2019), raising
Fig. 2. Dynamic public policy-cycle in the AI-enabled age. Source: Own ela- critical concerns about the sustainability of democracy and political
boration. pluralism.
The policy implementation stage could be deeply affected by the
5. Discussion development of the DPPC model. Here, disruptive changes could be part
of the generalization of intelligent automation, based on immediate and
This section discusses our approach and the results of the study. We efficient results with 24/7 services. This might increase public value
consider the previous sections and look at the future applications of the and citizens' attention (Valle-Cruz, 2019). Despite the replacement of
Dynamic Public Policy-Cycle (DPPC). We support the idea that the massive groups of public servants being a possible result of routine
potential of Artificial Intelligence (AI) for the public policy-cycle can activities being carried out by robots and intelligent machines; this si-
promote faster and more accurate analysis of the data. This data, in tuation can also empower the human factor. Hence, governments must
turn, forms part of the public agenda. AI has the potential to monitor invest in the training of public servants, as well as in public-sector in-
public policy throughout the cycle and to make real time amendments novation (Sun & Medaglia, 2019). Doing so increases focus on creativity
during the process. Nonetheless, the use of AI has two possible paths of and the most innovative part of their jobs.
development. It may lead to the use and appropriation of technology by Lastly, the policy evaluation stage could be revamped by the
only some interest groups, which cause exclusion and bias (Lazer, adoption of the DPPC model. AI can trace results at any stage of the
2016). Alternatively, it may lead to more informed decision-making, cycle. Enhanced facilities for monitoring public policies that AI offers
considering different scenarios to generate an agenda for the good of allow for the prediction and visualization of policy outcomes. These
society (Cath et al., 2018). emerging capabilities would facilitate decision-making and make it
A more dynamic public policy-cycle can be supported by the ana- more assertive. Feasible policies, which are designed to address com-
lysis of data through AI. Data can be monitored in real time and thus plex social issues, would be generated (Androutsopoulou et al., 2019).
evaluate the actions taken at each moment. This leads to greater effi- Therefore, evaluation might become an inner dimension of public
ciency, lower costs, and higher public value (Criado & Gil-Garcia, 2019; policy at any time. Additionally, evaluation wouldn't be dependent on
Valle-Cruz, 2019). However, relevant data must be in the hands of the will of public sector officials or political elected representatives.
governments, without violating privacy and confidentiality laws. Un- Despite the above-mentioned potential payoffs that AI might bring
ethical use of such data might lead to excessive control (Cath et al., about for the public policy-cycle, there are also some potential pitfalls
2018; Coglianese & Lehr, 2016; Wirtz & Müller, 2019). For this reason, that could be endorsed to the application of the DPPC model. At this
it is important to regulate AI and to reinforce data use policies (Vetrò point, we highlight the following. First, some countries that do not have
et al., 2019). Decision-making will be driven by data used for policy- the infrastructure or capacity for AI-adoption could be excluded. This
making (Giest, 2017; Höchtl et al., 2016). could increase the digital divide. Second, the design of intelligent al-
Our DPPC model has substantial benefits for decision-makers and gorithms could raise ethical concerns, since a biased design could lead
public policy-makers. To begin with, AI techniques used for data ana- to poor decision-making inherited by non-rational human behavior
lysis, as well as in the generation of robots and intelligent algorithms, (Wirtz & Müller, 2019), racism or social exclusion (Vetrò et al., 2019).
have the potential to make governments more agile (Mergel, 2017), Then, AI could intentionally or accidentally increase social and eco-
nomic inequalities (Thierer et al., 2017). Third, the design of AI must be

9
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

Table 4
Comparison between the traditional public policy-cycle and the dynamic public policy-cycle (DPPC).
Traditional DPPC

Agenda setting Activities are carried out to detect and select problems or The detection and selection of problems is automatic and preventive.
opportunities in society. Sometimes corrective measures are taken. This is due to the analysis of the Big Data, by means of AI techniques.
There are some problems or opportunities that are not part of public Prevention and data analysis can be invasive.
policy. Simulations, automation, and data processing help to solve problems
Only some members of the political community decide which issues that were previously unsolvable. There must be transparency and ethics
to address in the public agenda. in the design of intelligent algorithms so as not to generate biases or
Due to the complexity of public problems, there is no definitive results that affect vulnerable groups.
formulation. Citizens can participate in a more dynamic way and AI can give
recommendations for the improvement of the agenda. There may be too
many possible solutions. This fact in increases complexity.
Data analysis and simulations allow better solutions to be given or
certain policies to be discarded from the agenda because they do not
obtain the expected results.
Policy formulation and decision In practice, there are difficulties in identifying the objectives of The detection and generation of alternatives is easier, so the objectives
making organizations and programs, in addition to defining the objectives and goals will be set more quickly and accurately. Greater control
for future development. Sometimes it is slow and wrong. generated.
Already known alternatives are included in public policy and rarely Data analysis and simulations allow new opportunities to be explored,
are innovations or new alternatives explored. enhancing innovation. Technological innovation can displace humans
An assessment and comparison of alternatives are made in economic from their activities.
terms. AI techniques allow budget optimization, generating better options with
The choice of policies is in the hands of the decision-makers in the minimal cost. Departments may disappear or there may be layoffs of
government. people who are not needed, caused by AI automation.
Decision-makers rely on AI. Dependence on AI can be generated and
there is a risk of leaving decision-making to intelligent algorithms.
Policy implementation Financial and human resources are used to implementing the policy Humans are no longer required to perform routine activities. Results are
adopted. generated more efficiently, more cost effectively and 24/7.
The programmed sequence of actions depends on various actors AI allow sequences of actions to be performed unilaterally, with no dead
with different perspectives, attitudes, and interests. ends, but can lead to inflexibility.
Policy evaluation The evaluation may be delayed or even not carried out at all. AI and data analysis allow for the evaluation of public policy at each
stage, providing feedback for correction and adjustment cyclically and
automatically. This situation generates better, more strategic decision
making.
Reliance on AI can affect human intervention. The dynamics of the
public policy-cycle can only be carried out by AI.

Source: Own elaboration.

carried out transparently. We must remain acutely aware of data pro- machine learning, intelligent agents, big data, robotics, autonomous
tection and the right to privacy (Thierer et al., 2017). In order that there vehicles, data mining, chat-bots, pattern and facial recognition.
is certainty about how the decisions are made, citizens' distrust must be According to the literature, applications of AI in public administration
offset by assuring integrity of critical data. Finally, AI could lead to the are related to the analysis of data through intelligent algorithms that
replacement of a large number of decision-makers. They would be re- can foster social development, better decision-making, and forecasting.
placed by automation and algorithmic government scenarios (Engin & In addition, AI can be useful to improve citizen-oriented services.
Treleaven, 2019). We may become dependent on AI for decision- However, some risks could be foreseen. Risks include algorithmic dis-
making (Vetrò et al., 2019). crimination, opacity, social control, human replacement, and digital
divide.
This study also has some limitations that open up future research
6. Conclusions
avenues. First, the study is based on scientific literature, supported by
empirical studies. We might need to conduct a more comprehensive
In this article, we analyzed the public policy-cycle in the age of
analysis of the effects of AI on public policy by using different research
Artificial Intelligence (AI). The aim of this paper was to assess the
methods. Also, most of the case studies do not describe how the dif-
public policy-cycle by building on a new conceptual framework: the
ferent stages of the public policy-cycle were carried out, making diffi-
Dynamic Public Policy-Cycle (DPPC). Regarding the theoretical-con-
cult to devise the longitudinal implications of AI during the life-cycle of
ceptual review of AI literature, we performed a systematic analysis
the projects. Finally, the challenges for public policy in the age of AI are
based on 49 references. Doing so allowed us to study the concept in
relatively less studied than the potential benefits. Here, future studies
terms of public policy and administration research. On the other hand,
will be needed to understand different issues, including legal and moral
this work explored the foundations of the theoretical and practical re-
issues, ethical principles, legitimacy, transparency, workforce sub-
lationship between AI and public policy. Our analysis of AI at different
stitution through intelligent automation, predictions, or decision-
public policy-cycle stages is unprecedented and is of key empirical
making discretion.
importance. Likewise, our DPPC model configures a framework serving
To sum up, dynamism in the public policy-cycle may allow for the
as a cutting-edge reference in public policy and AI. Therefore, the DPPC
generation of better solutions to complex problems in a more efficient
is the most pervasive contribution of this article, as it displays the po-
and human-centered way. Each stage of the public policy-cycle can be
tential benefits (and pitfalls) of applying AI from the perspective of the
evaluated faster, more accurately and at a lower cost due to the massive
public policy process. Nonetheless, we expect future developments to
data processing and simulations that AI will bring from the onset. Data
be based on the dissemination of agile and dynamic perspectives in the
processing through AI techniques will generate valuable information
policy process.
for decision makers which, in turn, will become knowledge for the
One surmise of this article is that AI applications which are useful
formulation of public policy strategies. Our DPPC model is intended to
for the public policy-cycle will monitor data to discover patterns for
be applied in real settings to contribute to the latent changes in the
event prediction. Some AI techniques mentioned in the literature are

10
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

policy-process based on the application of agile methods and dynamics from https://www.jpl.nasa.gov/news/news.php?feature=6590.
in a context of growing implementation of AI techniques and applica- Greenemeier. (2014). Scientific American. Retrieved from https://www.
scientificamerican.com/article/smart-machines-join-humans-in-tracking-africa-
tions. Future studies will be required to evaluate the coming changes in ebola-outbreak/.
these processes and to appraise subsequent transformations of the Grothaus. (2019). Fast company. Retrieved from https://www.fastcompany.com/
policy-making process. 90324512/chinas-airport-facial-recognition-kiosks-should-make-us-fear-for-our-
privacy.
Hadden, S. G. (1989). The future of expert systems in government. Journal of Policy
Acknowledgements Analysis and Management, 8(2), 203.
Höchtl, J., Parycek, P., & Schöllhammer, R. (2016). Big data in the policy cycle: Policy
decision making in the digital era. Journal of Organizational Computing and Electronic
This study was partially supported by the Research Grant H2019- Commerce, 26(1–2), 147–169.
HUM 5699 (On Trust), Madrid Regional Research Agency and European Howlett, M., McConnell, A., & Perl, A. (2017). Moving policy theory forward: Connecting
Social Fund. multiple stream and advocacy coalition frameworks to policy cycle models of ana-
lysis. Australian Journal of Public Administration, 76(1), 65–79.
Howlett, M., Ramesh, M., & Perl, A. (2009). Studying public policy: Policy cycles and policy
References subsystems. 3. Oxford, England: Oxford university press.
Hwa, K. (2018). Three big trends for the future of public safety. Gov Insider.
Agarwal, P. K. (2018). Public administration challenges in the world of AI and bots. Public Jackson, P. C. (2019). Introduction to artificial intelligence. Courier Dover.
Administration Review, 78(6), 917–921. Jann, W., & Wegrich, K. (2007). Theories of the policy cycle. Handbook of Public Policy
Androutsopoulou, A., Karacapilidis, N., Loukis, E., & Charalabidis, Y. (2019). Analysis: Theory, Politics, and Methods, 125, 43–62.
Transforming the communication between citizens and government through AI- Janssen, M., & Helbig, N. (2018). Innovating and changing the policy-cycle: Policy-ma-
guided chatbots. Government Information Quarterly, 36(2), 358–367. kers be prepared!. Government Information Quarterly, 35(4), S99–S105.
Balke, T., & Gilbert, N. (2014). How do agents make decisions? A survey. Journal of Janssen, M., & Kuk, G. (2016). The challenges and limits of big data algorithms in
Artificial Societies and Social Simulation, 17(4), 13. technocratic governance. Government Information Quarterly, 33(3), 371–377.
Baraniuk, (2015). NewScientist. Retrieved from https://www.newscientist.com/article/ Joyner-Roberson, E. (2019). What do drones, AI and proactive policing have in common?
mg22830412-800-earthquake-artificial-intelligence-knows-where-damage-is-worst/. Published in SAS. Retrieved from https://www.sas.com/en/_us/insights/articles/risk-
Barth, T. J., & Arnold, F. (1999). Artificial intelligence and administrative discretion: fraud/drones-ai-proactive-policing.html.
Implications for public administration. American Review of Public Administration, Keynes, J. M. (2010). Economic possibilities for our grandchildren. Essays in persuasion
29(4), 332–351. (pp. 321–332). London, England: Palgrave Macmillan.
Birckland, T. (2016). An introduction to the policy process: Theories, concepts, and models of Lazer, D. (2016). The rise of the social algorithm. Does content curation by Facebook
public policy making. New York, NY: Routledge4. introduce ideological bias? Science, 348(6239), 1090–1091.
Bredereck, R., Chen, J., Faliszewski, P., Guo, J., Niedermeier, R., & Woeginger, G. J. Margetts, H. (2017). The data science of politics. Political Studies Review, 15(2), 201–209.
(2014). Parameterized algorithmics for computational social choice: Nine research Margetts, H., & Dorobantu, C. (2019). Rethink government with AI. Nature, 568(April),
challenges. Tsinghua Science and Technology, 19(4), 358–373. 163–165.
Bridgman, P., & Davis, G. (2003). What use is a policy cycle? Plenty, if the aim is clear. McKelvey, F., & MacDonald, M. (2019). Artificial intelligence policy innovations at the
Australian Journal of Public Administration, 62(3), 98–102. Canadian Federal Government. Canadian Journal of Communication, 44(2), 43–50.
Castro, D., & New, J. (2016). The promise of artificial intelligence. Center for Data Meijer, A. (2017). Datapolis: A public governance perspective on “smart cities”.
Innovation. Perspectives on Public Management and Governance, 1(3), 195–203.
Cath, C., Wachter, S., Mittelstadt, B., Taddeo, M., & Floridi, L. (2018). Artificial in- Mergel, I. (2017). Digital service teams: Challenges and recommendations for government. IBM
telligence and the “good society”: The US, EU, and UK approach. Science and Center for the Business of Government. Retrieved from http://kops.uni-konstanz.de/
Engineering Ethics, 24(2), 505–528. bitstream/handle/123456789/39164/Mergel_0-409608.pdf?sequence=3&
Cerrillo i Martínez, A. (2019). How can we open the black box of public administration? isAllowed=y.
Transparency and accountability in the use of algorithms. Revista Catalana de Dret Mergel, I., Gong, Y., & Bertot, J. (2018). Agile government: Systematic literature review and
Públic, 58, 13–28. https://doi.org/10.2436/rcdp.i58.2019.3277. future research.
Coglianese, C., & Lehr, D. (2016). Regulating by robot: Administrative decision making in Mikhaylov, S. J., Esteve, M., & Campion, A. (2018). Artificial intelligence for the public
the machine-learning era. The Georgetown Law Journal, 105, 1147. sector: Opportunities and challenges of cross-sector collaboration. Philosophical
Corvalán, J. G. (2018). Digital and intelligent public administration: Transformations in Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences,
the era of artificial intelligence. A&C-Revista de Direito Administrativo & Constitucional, 376(2128), Article 20170357.
18(71), 55–87. Newcombe, T. (2019). Quantum computing holds promise for the public sector. Published in
Craglia M, Annoni A., Benczur P., Bertoldi P., Delipetrev P., De Prato G., Feijoo C., government technology. Retrieved from https://www.govtech.com/products/
Fernandez Macias E., Gomez E., Iglesias M., Junklewitz H, López Cobo M., Martens Quantum-Computing-Holds-Promise-for-the-Public-Sector.html.
B., Nascimento S., Nativi S., Polvora A., Sanchez I., Tolan S., Tuomi I., Vesnic Alujevic Pencheva, I. (2018). Big data and AI—A transformational shift for government: So, what
L., Artificial intelligence—A European perspective, EUR 29425 EN, Publications O ce, next for research? Public Policy and Administration, 0(0), 1–21. https://doi.org/10.
Luxembourg, 2018, ISBN 978–92-79-97217-1, doi:https://doi.org/10.2760/11251, 1177/0952076718780537.
JRC113826. Pesapane, F., Volonté, C., Codari, M., & Sardanelli, F. (2018). Artificial intelligence as a
Criado, J., & Gil-Garcia, J. (2019). Creating public value through smart technologies and medical device in radiology: Ethical and regulatory issues in Europe and the United
strategies. From digital services to artificial intelligence and beyond. International States. Insights Into Imaging, 9(5), 745–753.
Journal of Public Sector Management, 32(5), 438–450. Poole, D. L., Mackworth, A. K., & Goebel, R. (1998). Computational intelligence: a logical
Criado, J. I., & Villodre, J. (2020). Delivering public services through social Media in approach. New York, NY: Oxford University Press.
European Local Governments. An interpretative framework using semantic algo- Russell, S. J., & Norvig, P. (2016). Artificial intelligence: A modern approach. Malaysia.
rithms. Local Government Studies doi:https://doi.org/10.1080/03003930.2020. Pearson Education.
1729750. Savaget, P., Chiarini, T., & Evans, S. (2019). Empowering political participation through
de Sousa, W. G., de Melo, E. R. P., Bermejo, P. H. D. S., Farias, R. A. S., & Gomes, A. O. artificial intelligence. Science and Public Policy, 46(3), 369–380. https://doi.org/10.
(2019). How and where is artificial intelligence in the public sector going? A lit- 1093/scipol/scy064.
erature review and research agenda. Government Information Quarterly, 36(4), Article Schuelke-Leech, B. A., Jordan, S. R., & Barry, B. (2019). Regulating autonomy: An as-
101392. sessment of policy language for highly automated vehicles. Review of Policy Research,
Desouza, K. (2018). Delivering artificial intelligence in government: Challenges and opportu- 36(4), 547–579.
nities. IBM Center for the Business of Government. Soe, R. M., & Drechsler, W. (2018). Agile local governments: Experimentation before
Dobell, R., & Zussman, D. (2018). Sunshine, scrutiny, and spending review in Canada, implementation. Government Information Quarterly, 35(2), 323–335.
Trudeau to Trudeau: From program evaluation and policy to commitment and results. Soumik, R. (2018). Indian railways leapfrogs into the era of AI. Techwire Asia. Retrieved
Canadian Journal of Program Evaluation, 32(3). from https://techwireasia.com/2018/12/indian-railways-leapfrogs-into-the-era-
Dunn, W. N. (2012). Public policy analysis (5th ed.). New York, NY: Routledge. of-ai/.
Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T., ... Galanos, V. Sun, T. Q., & Medaglia, R. (2019). Mapping the challenges of artificial intelligence in the
(2019). Artificial intelligence (AI): Multidisciplinary perspectives on emerging chal- public sector: Evidence from public healthcare. Government Information Quarterly,
lenges, opportunities, and agenda for research, practice and policy. International 36(2), 368–383.
Journal of Information ManagementArticle 101994 In press. Tanimoto, S. L. (1987). Elements of artificial intelligence. Computer Science Press.
Engin, Z., & Treleaven, P. (2019). Algorithmic government: Automating public services Thierer, A. D., Castillo, A., & Russell, R. (2017). Artificial intelligence and public policy.
and supporting civil servants in using data science technologies. The Computer George Mason University. Retrieved from https://www.mercatus.org/publications/
Journal, 62(3), 448–460. technology-and-innovation/artificial-intelligence-and-public-policy.
Floreano, D., & Mattiussi, C. (2008). Bio-inspired artificial intelligence: Theories, methods, Valle-Cruz, D. (2019). Public value of e-government services through emerging technol-
and technologies. MIT Press. ogies. International Journal of Public Sector Management. https://doi.org/10.1108/
Giest, S. (2017). Big data for policymaking: Fad or fasttrack? Policy Sciences, 50(3), IJPSM-03-2018-0072.
367–382. Valle-Cruz, D., Alejandro Ruvalcaba-Gomez, E., Sandoval-Almazan, R., & Ignacio Criado,
Good. (2016). Jet Propulsion Laboratory, California Institute of Technology. Retrieved J. (2019). A review of artificial intelligence in government and its potential from a
public policy perspective. 20th annual international conference on digital government

11
D. Valle-Cruz, et al. Government Information Quarterly 37 (2020) 101509

research (pp. 91–99). ACM. Monday, Information Polity, Local Government Studies, International Journal of Public
Van der Voort, H. G., Klievink, A. J., Arnaboldi, M., & Meijer, A. J. (2019). Rationality and Administration, or International Journal of Public Sector Management(among others). His
politics of algorithms. Will the promise of big data survive the dynamics of public research interests include algorithmic governance and artificial intelligence in the public
decision making? Government Information Quarterly, 36(1), 27–38. sector; open government and policies for transparency, participation and public innova-
Vetrò, A., Santangelo, A., Beretta, E., & De Martin, J. C. (2019). AI: From rational agents tion; and social media and big data in government.
to socially responsible agents. Digital Policy, Regulation and Governance. Retrieved
from http://www.antoniovetro.it/authorsversion/journals/2019-dprg-ai.pdf. Rodrigo Sandoval-Almazán (PhD) is Associate Professor at the Political Sciences and
Wirtz, B., Weyerer, J., & Geyer, C. (2018). Artificial intelligence and the public Social Sciences Department in the Autonomous University of the State of Mexico, in
sector—Applications and challenges. International Journal of Public Administration, Toluca City. Dr. Sandoval-Almazan is the author or co-author of articles in Government
1–20. https://doi.org/10.1080/01900692.2018.1498103. Information Quarterly, Information Polity, First Monday, Government; Journal of Information
Wirtz, B. W., & Müller, W. M. (2019). An integrated artificial intelligence framework for Technology for Development; Journal of Organizational Computing and Electronic Commerce;
public management. Public Management Review, 21(7), 1076–1100. International Journal of E-Politics IJEP. His research interests includes: artificial in-
Zuboff, S. (2019). The age of surveillance capitalism. London, England: Profile Books. telligence, social media in government. Public innovation, digital government and open
government.
David Valle-Cruz (PhD) is Professor at the Computing Engineering Department in the
Autonomous University of the State of Mexico, in Toluca City. Dr. David Valle-Cruz is the Edgar A. Ruvalcaba-Gomez is a Research Professor at the Universidad de Guadalajara
author or co-author of articles in Information Polity, First Monday, International Journal of (UDG), at the Center for Economic and Administrative Sciences, with affiliation to the
Public Sector Management, International Journal of Public Administration in the Digital Age, Research Institute in Public Policies and Government. Ph.D. in Law, Government and
and Digital Government: Research and Practice. His research is related to Applied Artificial Public Policies from the Universidad Autónoma de Madrid, Spain. Edgar conducts re-
Intelligence and Emerging Technologies in Government. search related to Open Government, Transparency, Citizen Participation, Open Data,
Digital Government, Artificial Intelligence, Corruption, and Public Innovation. He has
J. Ignacio Criado (PhD) is an Associate Professor / Senior Lecturer in Political Science been a visiting researcher at Trinity College in Dublin, Ireland, and at the Center for
and Public Administration, Department of Political Science and International Relations, at Technology in Government at the University of Albany, New York, USA. He has been
Universidad Autónoma de Madrid. His articles have been published in leading journals, Researcher for the Open Government Partnership in the program of subnational gov-
including Government Information Quarterly, Social Science Computer Review, First ernments. Coordinator of the Open Government Academic Network in Mexico.

12