Exploratory Insights on Artificial Intelligence for Government in Europe

Abstract

There is great interest to use artificial intelligence (AI) technologies to improve government processes and public services. However, the adoption of technologies has often been challenging for public administrations. In this article, the adoption of AI in governmental organizations has been researched as a form of information and communication technologies (ICT)–enabled governance innovation in the public sector. Based on findings from three cases of AI adoption in public sector organizations, this article shows strong similarities between the antecedents identified in previous academic literature and the factors contributing to the use of AI in government. The adoption of AI in government does not solely rely on having high-quality data but is facilitated by numerous environmental, organizational, and other factors that are strictly intertwined among each other. To address the specific nature of AI in government and the complexity of its adoption in the public sector, we thus propose a framework to provide a comprehensive overview of the key factors contributing to the successful adoption of AI systems, going beyond the narrow focus on data, processing power, and algorithm development often highlighted in the mainstream AI literature and policy discourse.

Keywords

artificial intelligence public sector innovation adoption AI-enabled innovation digital transformation

Introduction

Advances in machine learning have led to increased interest in artificial intelligence (AI) over recent years by all sectors of society, expecting it to become the key technology driving the next industrial revolution (Chui et al., 2018). Public organizations have also recently caught up on the promise of AI and started coordinating their efforts to use it to improve government administrative processes and services to the citizens (Mehr, 2017). AI technologies, in fact, hold the potential of improving the effectiveness, efficiency, and personalization of public services (Mehr, 2017). Therefore, there is a great interest in understanding how it could be used in the public sector and what value could be gained from its adoption. Nevertheless, previous research on eGovernment has shown that there are considerable challenges for public sector organizations adopting innovations, especially when it involves information and communication technologies (ICTs; Agarwal, 2018; De Vries et al., 2016; de Vries et al., 2018).

We still understand very little about how and why emergent technology—such as AI—is used within government (Kankanhalli et al., 2019). As the expectation is that AI will be used in more fundamental governmental processes (Engstrom et al., 2020), we believe that researching AI use could give valuable insights for policy makers to help understand which factors are more likely to support its adoption in the public sector.

To this end, our analysis follows an exploratory case study research design on three different applications of AI within the European public sector, addressing our main research question: “Which antecedents of public sector innovation enable the adoption of AI in public administrations in the European Union?” Such exploratory research is well suited to gain a broad understanding of emerging social phenomena such as AI, as it allows for a more in-depth understanding of the factors contributing to the adoption of innovations (Meijer, 2015).

The article aims to contribute to the academic debate on how antecedents of public sector innovation, which are already established in the research field, could be extended in order to better understand under which conditions AI innovations develop. In fact, the public sector is often left out of scope in most AI-related research as a recent review highlighted (Sousa et al., 2019). Only recently, more researchers have started exploring the actual applications of AI in the public sector, which highlights that the use of AI is less sophisticated or radical as anticipated as there are numerous barriers hindering its use (Engstrom et al., 2020).

The article is structured as follows: After an introduction of how AI is defined in this research, the different factors contributing to innovation in the public sector are discussed. A brief overview of our methodological approach is then presented, and the analysis of three cases of AI in use in different public administrations in Belgium, Estonia, and the Netherlands are described from the perspective of public sector innovation theory. The article ends with discussion of the findings and a conclusion outlining future research directions and implications for policy in the European context.

Literature Review

Conceptualizing AI in the Public Sector

Due to recent advances in computing power and algorithms, and the explosion of data availability, many applications using machine learning have been introduced in different areas of the economy and our daily life. This made the term AI revamped, and it has been associated with several new products that often use the terms big data, machine learning, or deep learning interchangeably with AI (Katz, 2017; Makridakis, 2017).

As a consequence of the lack of a clear conceptualization, however, researchers refer to AI in very different ways (Krafft et al., 2019). Some describe AI as software able to do intelligent tasks (Russel & Norvig, 2016; Sousa et al., 2019). Others prefer to research AI as a tangible technology rather than a goal-oriented tool (Scherer, 2016). In this article, we focus on the adoption of AI applications or ICT systems with capabilities, such as perception, learning or understanding, commonly regarded as human-like (Wirtz et al., 2019), and not so much in the development of models using machine learning (Desouza et al., 2020).

Despite the lack of a clear conceptualization on AI, scholars argue that AI technologies are able to provide value and benefits to government organizations in numerous ways (Alexopoulos et al., 2019; Eggers et al., 2017; Stone et al., 2016; Sun & Medaglia, 2019). AI applications are expected to be well suited to tackle common governmental problems in resource allocation, gaining insights from large databases, a shortage of experts to tackle certain problems, doing many repetitive procedural tasks and handling diverse data (Mehr, 2017), tackling corruption (Lima & Delen, 2020), and achieving social development goals (Vinuesa et al., 2020), although limited empirical proof is available. As an example, Chatbots are seen to be able to improve the communication between citizens and government agencies (Androutsopoulou et al., 2019) but might not live up to these expectations due to existing eGovernment challenges (van Noordt & Misuraca, 2019).

Much attention has also been given to possible negative effects of using algorithms within public sector organizations, as it creates opaque decision-making processes (Craglia et al., 2018; Pasquale, 2015; Preece et al., 2018), challenges in accountability and trust in AI-enabled decisions (Burrell, 2016), and risks to privacy due to sensitive, granular, and in-depth data collection practices (Floridi, 2017; Mittelstadt et al., 2016). In addition, the frequently mentioned risk of discrimination due to bias is another possible negative effect that attracted much attention in research and policy (Barocas & Selbst, 2016; Veale et al., 2018). These different perspectives on the value and risks of using AI are likely to influence the choices to adopt this technology within the public sector and the acceptance of innovation in society.

Understanding the Antecedents of Public Sector Innovation

In fact, one of the main challenges for governments is to adopt and use ICT innovations in their operations (Kamal, 2006; Potts & Kastelle, 2010). Hence, numerous scholars attempted to understand the conditions that make it more likely for innovation to occur in the public sector, defining the factors that act as a barrier or as a driver (also referred to as antecedents), which influence the adoption and diffusion of innovations in public organizations. Following the seminal work of Borins (2000) and Rogers (2010), a coherent framework has been developed by De Vries et al, (2016).

In this article, we follow this approach with the aim to unravel the “black box” of AI adoption within public organizations, considering it as a form of public sector innovation. This is in line with recent research from Schedler et al. (2019), stating that barriers to adoption of innovations in government remain the same, no matter what kind of innovation is introduced. We thus consider the following antecedents influencing the adoption of AI in the public sector: environmental, organizational, innovation-related, and individual.

Environmental antecedents

The working environment in which public organizations operate has significant effects on their innovative capacities. Most innovations are coming from a specific local context where numerous pressures from the environment, such as public opinion, media, or political activity, result in innovation (De Vries et al., 2016; Sørensen & Torfing, 2011). The networks in which organizations operate are also an important driver for innovation. Public organizations involved in frequent contact with other innovative organizations usually take over their norms and practices, as they are perceived to be legitimate or better, a process called mimetic isomorphism, resulting in organizations developing more innovative practices themselves (De Vries et al., 2016; Hinings et al., 2018).

In addition, when organizations in a network perceive that they are mutually dependent on each other, they are more likely to explore new innovations by sharing organizational resources (Bekkers et al., 2013; Bertot et al., 2016). These networks could also involve private vendors. Private sector organizations have been argued to play a major role in promoting and consulting the adoption of eGovernment services (Jun & Weare, 2011). Lastly, regulation is generally seen as a hindering factor to facilitate innovation, but it may also be a driver for innovations to occur as a result of the need to deal with imposed restrictions (De Vries et al., 2016).

Organizational antecedents

At the organizational level, there are numerous structural and cultural factors contributing to the adoption of innovations within the public sector. The more influential antecedent is the disposal of appropriate organizational resources (De Vries et al., 2016). In order to adopt innovations, there should be an adequate amount of time, money, and people available, something that is often limited in the public sector. Having an inadequate budget to adopt AI is in fact a huge barrier in many public sector organizations (Wirtz et al., 2019). Naturally, for adopting ICT-enabled innovations, there should also be enough staff available with appropriate competences (Cinar et al., 2018; Meijer, 2015). However, the demand for experts with AI skills is extremely high, whereas their availability is low. Getting the necessary expertise to develop and adopt AI might, therefore, be challenging and expensive in the short term (Centre for Public Impact, 2017; Susar & Aquaro, 2019).

Another often-mentioned antecedent is the participation of relevant stakeholders in the development process as it makes adoption of the innovation more likely (Lewis et al., 2018). If end users do not have the skills to use the innovative product or service as intended, there should be training and support available to increase acceptance (de Vries et al., 2018).

There should also be a supportive technical infrastructure with enough processing power, storage, bandwidth, and connectivity for digital innovations to emerge (Bertot et al., 2016). For AI applications using data from different sources, it is necessary that the systems are interoperable, including the capacity for different data to readily work with each other across different systems (Kankanhalli et al., 2019). This requires sufficient expertise within the organization on effective data management, complemented by technical skills, as the data used for AI need to be cleaned, integrated, structured, and secured (Harrison et al., 2019).

A lack of management support and/or credible leadership with a vision for integrating new solutions in the administrative processes is another frequently mentioned barrier (Meijer, 2015). The organizational culture could stimulate the adoption of innovations, possibly when accompanied with financial incentives or other rewards (De Vries et al., 2016).

Innovation-related antecedents

Innovations need to be perceived as “value adding” by all stakeholders involved in the adoption process. Innovations should also be regarded as easy to use and to experiment with and compatible with the organizational values and, to a certain extent, historical experiences, in order to facilitate their adoption (Cinar et al., 2018; De Vries et al., 2016). This is of particular importance for AI-enabled innovations as the potential for radical innovation is high and their adoption could disrupt existing processes and in turn change previous administrative practices that with time may have become established cultural norms, considerably.

Individual-related antecedents

Lastly, there are specific factors regarding the role of individuals involved in the process of innovation that is crucial for its adoption. Frequently, it is argued that creative leadership is needed in order to overcome the previously mentioned environmental or organizational barriers. An individual within an organization, no matter their position in the hierarchy, can be seen as the informal leader of and an important factor in innovation (De Vries et al., 2016). It is in fact often the case that within an innovation process, there is an individual who spots the potential of a new technology and persuades their colleagues to adopt it (Kamal, 2006).

Although the environmental and organizational antecedents play the largest role in enabling different forms of innovation (De Vries et al., 2016), it is the combination of all of them which provides a view of how and why innovation in the public sector occurs.

Methodological Approach

To answer our research question, we conducted an exploratory multiple case study on three different adoptions of AI within public sector organizations in EU countries. Since AI within the public sector has received little attention from scientific research, following Yin (2018), the exploratory case study design allows an early examination of this relatively new phenomenon, with the aim to test current theories and generate new ones (Flyvbjerg, 2006).

The use of multiple case studies is also expected to offer more replicable, reinforced, and robust findings that help to provide a more generalizable contribution to academic research (Baxter & Jack, 2008). In our research, however, we consider it crucial that different AI technologies—as well as the context in which they operate—are comparable with each other, as the term encompasses multiple technologies that might not be alike. While the analytical framework can thus be applied universally, we acknowledge that historical, institutional, and cultural elements are not captured in this study, influencing the findings.

To facilitate the case study selection, the research built upon the cases gathered by the European Commission’s Joint Research Centre’s AI Watch and the AI Alliance. The AI Watch has been set up to monitor the AI landscape in the European Union (EU) in both the private and public sectors (European Commission, 2018). Most of these cases are self-reported by the Member States as being AI. As summarized in Figure 1, three of the 82 cases gathered in the period February–May 2019 have been selected based on the following criteria: the country adopting the technology, accessibility and availability of information about the potential candidate case, language and quality of the documentation available online.

Figure 1.

Case studies selection process.

The cases selected are the SATIKAS system in Estonia which use AI technologies to check land mowing, the predictive system for day-care services inspection used in the Flemish Child and Family Agency in Belgium, and AmberScript, an automatic transcription tool to provide subtitles for video recordings of political council meetings used in the Province of Gelderland in the Netherlands.

Once the cases have been selected, the analysis of the AI applications and their adoption has been done in three steps, through desk research, interviews, and validation.

First, a document analysis was conducted on available material online or provided by other means to gain a brief overview of the main purposes of the AI application. SATIKAS, for example, was presented by the Estonian government during the Tallinn Digital Summit 2019 and has been described in policy reports (Network of European Regions Using Space Technologies [NEREUS], 2018; Organisation for Economic Co-operation and Development [OECD], 2019). This document analysis was conducted through a snowball method using both Google Scholar and Google search engine.

To complement the analysis, four semistructured interviews with the persons responsible for introducing the innovation in their organization and other experts were carried out. The interviewees were either the project managers or main contact person regarding each project. One respondent from the SATIKAS case preferred to answer questions by email while most interviews were conducted remotely, apart from the interview with AmberScript which was in person. Each interview lasted for an average of 30 min. The questions were customized based on the context of the case. They, however, remained comparable as they were based on the antecedents of public sector innovation which served to structure the interview.

The interviews have been used to complement and validate the information found in other data collection methods but were considered crucial in understanding how and why the process of innovation occurred. If there was any need of clarification, additional email correspondence with the respondents followed. Moreover, the case analysis has been reviewed by the respondents themselves in order to correct possible misinterpretations by the authors.

However, while these interviews have been insightful, there is a risk of bias that the project managers see the innovation as a great success, a perspective that might not be shared by other actors. The lack of additional interviews with other civil servants working with the AI application or users of the systems limits some of the findings regarding the perceived value of the innovation by different stakeholders.

Finally, a focus group with experts who are part of the AI Watch took place at the European Commission’s Joint Research Centre in June 2019, to validate the results of the case study analysis and place them within the context of the research, as well as discussing future research and policy implications.

Case Studies on the Use of AI in Government

In this section, we briefly illustrate the three cases of use of AI in public administrations we analyzed in Estonia, the Netherlands, and Belgium. These cases are illustrative and served to start building the knowledge base for future comparative analysis and more in-depth research.

SATIKAS

In the Estonian Agricultural Registers and Information Board (ARIB), the system called SATIKAS¹ uses satellite data coming from the European Copernicus Programme to control automatically, using AI technologies, whether mowing has taken place on the Estonian grasslands (Tartu Observatory, 2019). With increasing labor costs in Estonia, it was becoming more and more expensive to have checks performed by field inspectors (NEREUS, 2018) leading to the need to reduce the number of visits while preventing farmers from not keeping up with the subsidy requirements (Bleive & Voormansik, 2016).

SATIKAS was developed gradually as part of applied research conducted jointly by ARIB, CGI, and the Tartu Observatory (Bleive, 2017), after enthusiasts met together informally in 2011. The system uses the deep learning methods recurrent and convolutional neural networks for the analysis of the satellite data (Respondent Tartu Observatory, personal communication, 2019). Data from the Sentinel-1 radar and Sentinel-2 optical satellite images, together with ground reference data of some farmer’s fields, historical inspection logs of ARIB, and meteorological data from the Estonian Weather Service were used to conduct the analysis (Commission, 2017; NEREUS, 2018).

Despite the general interest for trying out new technologies, some senior officials at ARIB were not initially very hopeful about the success of the project (Respondent Tartu Observatory, personal communication, 2019). Some staff feared the creation of a “Big Brother State,” while others feared that their jobs might disappear due to the introduction of the technology (Estonian Agricultural Registers and Information Board, personal communication, April 26, 2019)). Nevertheless, the project was allowed to continue as an experiment to see what was possible (Respondent Tartu Observatory, personal communication, 2019).

The project gained funding through the European Regional Development Fund to assist the development of public services with ICT (NEREUS, 2018). Different resources and capabilities from all the various organizations involved were used in order to develop SATIKAS. This included machine learning expertise from the Institute of Computer Science of the University of Tartu and the Software Technology and Applications Competence Centre. It must be noted that the technological infrastructure used to develop SATIKAS has changed over time once the project grew. The technological platform of ARIB became in fact insufficient to store and handle reliably the huge data volumes involved, leading to the use of the infrastructure of the Environmental Agency of the State Service (Respondent Tartu Observatory, personal communication, 2019).

Having access to high-quality data—from both the ground and the satellite imagery—has been an important condition for the SATIKAS system to succeed. In the beginning of the project, the data supply from Copernicus was not always of high quality due to changing formats and duplicate images, leading to a lot of work when managing the quality of the data sets (Respondent Tartu Observatory, personal communication, 2019).

Civil servants within ARIB using the SATIKAS system were partially involved during the system development, providing verification data and assessing the outputs generated. They also received training to understand how it works in order to gain trust in its results (Respondent Tartu Observatory, personal communication, 2019). After the development of the system, the field inspectors realized that fears about job losses were inappropriate as their man power was still required (Estonian Agricultural Registers and Information Board, 2019). Other stakeholders started to see more value in the system once the first results were satisfactory.

Predictive System Day-Care Services

In 2014, the Flemish Agency for Child and Family (Kind en Gezin) in Belgium started a pilot project to use advanced data analytics to create a predictive model to detect day-care services that require further inspection (Bongers et al., 2018). The Child and Family Agency does not carry out the inspections itself but works together with the regional Health Care Inspectorate of the Department of Welfare, Public Health and Family (European Commission, 2019). However, there is limited capacity to conduct inspections, which led to the need of optimizing the inspection process (Bongers et al., 2018). Data mining was thus introduced as a contribution to inspection practices based on the experience and reasoning of the staff in the Healthcare Inspection systems, in turn making them more accurate (Respondent Kind en Gezin, personal communication, April 4, 2019).

The predictive system for the Child and Family agency is based on a supervised machine learning method. A logistic regression and XGBoost were used, as these methods had the best results in early testing (Respondent Kind en Gezin, personal communication, April 4, 2019). In order to develop the predictive system, data found in the internal data warehouses and data from the Health Inspectorate were analyzed (Bongers et al., 2018). The current predictive model was released in 2017, but its development was inspired by a previous attempt. In fact, at the end of 2013, Child and Family worked together with IBM to design a predictive model to indicate which kind of day-care services should be subject to inspection. However, due to legal and organizational changes in 2014, the previous model was no longer applicable to the new situation (Bongers et al., 2018).

During the new system’s development, the agency cooperated closely with the Data Science team of the Department of Welfare, Public Health and Family because of their expertise in text mining (Bongers et al., 2018). In addition, there was a strict collaboration with the Health Inspectorate, as they provided the data used in the system. A consultancy played a supporting role by giving additional expertise in the use of RStudio (Respondent Kind en Gezin, personal communication, April 4, 2019).

Despite general enthusiasm for innovations and technology within the agency, the AI application had to be seen as a tool to empower and support workers rather than replacing them or checking if they are doing their work well (Respondent Kind en Gezin, personal communication, April 4, 2019). Therefore, staff was involved as much as possible throughout the project (Bongers et al., 2018). The combination of both showing statistical proof of the validity of the system, and an emphasis on supporting human workers, rather than replacing them, further improved the acceptance of the system by the end users (Respondent Kind en Gezin, personal communication, April 4, 2019).

While a small part of the budget of Child and Family could have been made available for IT data science projects, employees worked on it in their spare time, with limited resources as the project (in its initiation at least) was initiated as an experiment so to avoid possible resistance (Respondent Kind en Gezin, personal communication, April 4, 2019). The availability of enough trustable and high-quality data was an important factor in enabling the adoption of the AI tool. This is crucial for any data mining or AI project, although some additional tasks were required in order to comply with the General Data Protection Regulation (GDPR; Respondent Kind en Gezin, personal communication, April 4, 2019).

In addition, to gain and maintain trust of users in the recommendations provided by the AI, and to ensure accuracy and reliability of data, there is a need for constant maintenance and improvement of the model. It was thus suggested that not only would poor data maintenance lead to a possible decrease in model accuracy but also a reduction in the trust of other data projects.

Amberscript

In 2018, the Province of Gelderland in the Netherlands adopted an automatic transcription tool to provide subtitles for video recordings of political council meetings. Before this, the organization’s clerks usually made the transcriptions or summaries of the meetings themselves. Sometimes, other external parties were contracted to do the transcriptions, but it could take over 3 months before they would become available (Respondent Province of Gelderland, personal communication, May 6, 2019).

The automated transcription of the meetings is made possible by the AI tool AmberScript. This software uses speech recognition technology in order to interpret and convert the words spoken in audio files into text used for summaries and subtitles. At the moment, apart from the Province of Gelderland, AmberScript is already in use in around 60 municipalities in the Netherlands (Respondent AmberScript, personal communication, April 17, 2019).

In 2018, in order to provide automatic transcriptions of the meetings, AmberScript has partnered with WebCast, a private organization which provides a platform for over 150 provincial and municipal public bodies to host video and audio recordings of their meetings and to make them available to the public (Respondent AmberScript, personal communication, April 17, 2019; WebCast, 2019). For organizations already using WebCast’s services, the automated transcriptions are an additional functionality which can be purchased on demand.

This partnership was needed in order to develop a specific speech recognition system since WebCast allowed access to many existing audio and video files from previous political meetings (Respondent AmberScript, personal communication, April 17, 2019). The speech recognition model was trained using a combination of the audio and text databases that had been filled with handmade or previously available transcripts of meetings. In order to train the speech recognition model for high-accuracy transcription, a data set of more than 1,000 hr of audio with transcriptions was compiled, cleaned, and processed into the AI model (Webcast, 2019). The availability of all the recordings of the council meetings was one of the key factors in the development of the system (AmberScript, personal communication, April 29, 2019).

Webcast has been providing services for the Province for more than 12 years and, due to this long-standing relationship, notions of shared trust and joint commitment can be seen as drivers for innovations to be adopted (Respondent AmberScript, personal communication, April 17, 2019; Respondent Province of Gelderland, personal communication, May 6, 2019). However, an early version of the automated transcriptions did not meet the expectations, so there were additional discussions with Webcast on how to improve the technology. Later, adjustments and pilots were proven to be satisfactory, and the decision was taken to integrate the use of the system in the transcription process of the council meetings (Respondent Province of Gelderland, personal communication, May 6, 2019).

An important factor that led to the adoption of AmberScript was a change in Dutch law due to the 2016 European Directive on digital accessibility,² which required websites and mobile apps of governmental institutions to be accessible for citizens with handicaps (Ministerie van Binnenlandse Zaken en Koninkrijksrelaties, 2017; Respondent Province of Gelderland, personal communication, May 6, 2019). While this law was considered important, it was not yet in force and so the technology did not need to be fully adopted. Nevertheless, the Province wanted to stay ahead of the deadline as they agreed with the aim of the new law and saw it as an obvious mechanism to make their services as accessible as possible (Respondent Province of Gelderland, personal communication, May 6, 2019).

The other main reason for the adoption of the automatic subtitle system was the general organizational culture and practice of adopting innovation which facilitate improvements to the services of the Province. Strong beliefs were expressed that governmental institutions should keep innovating when appropriate, staying aware of what is happening on the technological market and being open to trying innovative solutions (Respondent Province of Gelderland, personal communication, May 6, 2019).

The clerks who are now using the system see it as adding considerable value since it saves them a lot of time and effort. The use of the tool has allowed them to focus on other tasks that provide more value to citizens, though they still have to check the translations for possible mistakes (Respondent AmberScript, personal communication, April 17, 2019). The system is easy to use and there was no training required for its introduction, although there was a collaboration with Webcast to help using the system (Respondent Province of Gelderland, personal communication, May 6, 2019) which shows the importance of innovative public–private partnerships and procurement models for promoting adoption of AI in the public sector.

Findings and Discussion

The analysis of case studies of ICT-enabled innovations from the perspective of the antecedents that act as drivers and/or barriers to adoption helps us to better understand what factors can hamper the full application of AI’s potential and those that can, instead, facilitate its adoption on a wide scale across the public sector.

For this reason, we discuss the antecedents of innovation introduced in the literature review as they emerged in the analysis of the case studies. In doing so, we aim to discover whether any other AI-specific factor arises.

Environmental Antecedents

For most innovations in the public sector, the context in which the organization functions has a significant effect on the capability to introduce and likelihood to adopt innovations, as seen as in Table 1. Based on the case analysis, the role of networks has proven vital in the development of AI. This shows that the local environment in which the institution operates influences their adoption of AI.

Table 1.

Environmental Antecedents.

Environmental Antecedents	SATIKAS	Child and Family Predictive System	AmberScript
Local pressure	Rising labor costs and regulation contributed indirectly	Pressure to improve inspection capabilities	New upcoming regulation created pressure for the organization to adopt the system to comply
Networks	Great contribution to awareness, development, and adoption	Crucial for initiation and development of the system	Crucial for awareness, development, and adoption of the system
Private vendors	Minor role	Minor role, major role in previous project	Major role in promoting, development, and adoption
Isomorphism	No isomorphism as SATIKAS was regarded as a first mover in the field	Mimetic isomorphism through conference	It is used by other organizations, but this had no recorded influence on the adoption
Regulation	Regulation increased the need for the system, but it is not hindering development or adoption	General Data Protection Regulation had an effect on handling of data	Upcoming regulation stimulated adoption of the system

These networks not only include other public sector organizations but also private actors. In each of the cases investigated, the roles of the private companies differed, although their involvement was as a partner in the development of the AI tools rather than simply as a vendor. This shows the benefit of having collaborative partnerships between public and private sector actors, including the role of both professional and personal contacts in forming them.

In addition to the importance of networks, all adoptions were influenced by some form of environmental trigger, whether indirectly (e.g., rising labor costs and more legal requirements) or directly (e.g., regulation on digital accessibility). These pressures from the environment enabled making the usage of AI a more valuable choice, although this was not a direct consequence of this environmental pressure. However, once the AI technology became a legitimate alternative to tackle these environmental pressures, such as the case of AmberScript, adoption seems to be more likely to occur successfully.

Organizational Antecedents

Both structural and cultural organizational factors have frequently been mentioned to be an important factor in the adoption of innovations within the public sector. The cases analyzed confirm that there are numerous cultural and organizational factors that have contributed to the adoption of the AI systems within the governmental organizations examined, summarized in Table 2.

Table 2.

Organizational Antecedents.

Organizational Antecedents	SATIKAS	Child and Family Predictive System	AmberScript
Organizational resources and funding	Sufficient but high cost of developers limit progress. Funding through European Union budget	Sufficient to develop and adopt the system. But lack of staff resources increased time	Specific budget for innovations with service provider (Webcast)
IT resources	Shared IT-resources and capabilities	IT resources and capabilities shared or present in-house	IT resources outsourced
End user participation	Partially	Close collaboration by giving multiple presentations	Close collaboration in testing and piloting
Training	Provided to enable trust in the system	No training on the use of the system	No training
Management support	Sceptical at first	Positive	N/A
Organizational culture	Innovative; eager to do more with less	Positive toward technology and innovation	Positive toward technology to improve access to information for citizens and making government open
Incentives	No financial rewards, initiation out of personal interest	No financial rewards, initiation out of personal interest	No financial rewards

Insights from the cases also demonstrated that it is challenging to find expertise in AI. None of the organizations had (all) the necessary expertise in-house to develop the systems. Rather, there were numerous organizations contributing to the development or adoption of AI with varying levels of involvement.

In all the cases, gradual improvements on the system were required, as it was likely that mistakes would be made and poor performance is expected, at least at the start of the project. This shows, in part, the “learning” character of the technology, but it also means that some stakeholders may need to be more patient to wait for positive results. The role of senior management is vital in shaping such a culture. Even when senior officials might be critical of using AI technologies in administrative processes and public services, experiments should be allowed to discover what benefits a technology is able to bring to established practices and what effects it may have on the organizational performance.

Moreover, end users, in particular civil servants working with the applications, should be consulted and their feedback taken into consideration. As it emerged from the interviews, due to the acceptance from users, the added value of the AI applications is likely to be higher.

Innovation Antecedents

While the environmental and organizational antecedents play a significant role in the adoption of AI, there are some specific attributes connected to AI technologies, themselves, that have led to their adoption, as seen in Table 3.

Table 3.

Innovation Antecedents.

Innovation Antecedents	SATIKAS	Predictive System	AmberScript
Perceived Value	Valuable addition to work more effectively	Valuable contribution to support staff	Valuable as it reduces work burden
Compatibility with organizational values	In line with organizational values to do more with less	Improvements of field inspection in line with organizational tasks	Fits organizational value of inclusiveness of public services
Ease of Use	Data visualization of results makes the system easy to use	Data visualizations of results make the system easy to use	Strong collaboration with end users to ensure ease of use
Security and privacy concerns	No sensitive data used, so no security or privacy concerns	No sensitive data used, so no security or privacy concerns	No privacy concerns, as it regard public meetings

In all the cases, AI was considered to provide value according to the project leaders. However, results from the analysis of the cases show that the perception of value is “dynamic” and may differ between various stakeholders.

For instance, at the start of the SATIKAS project, many civil servants did not see any value in the system. After showing and discussing results, the perception of the value increased. Thus, it should be explained clearly that the introduction of an AI system is an augmentation to the work of civil servants rather than a replacement. In this respect, the dynamics of the process of reducing organizational resistance toward AI-systems are not yet well understood and should be further investigated enlarging the empirical base with primary survey data.

Individual Antecedents

In the different AI systems studied, we have identified certain individuals who have played a significant role in their adoption. For example, in the SATIKAS case, one representative of ARIB was mentioned as having played a significant role in its adoption, being referred to as the “soul of the development” (Respondent Tartu Observatory, personal communication, 2019). One aspect to underline related to the individuals involved in the adoption of AI was their personal interest, to the extent that some were working on the projects in their free time. This shows a strong inner motivation to do extra activities in their job such as experimenting with new technologies such as AI.

AI Antecedents

In addition to the antecedents typical for public sector innovation, we assumed that there are specific factors summarized in Table 4 that are crucial for AI adoption, and, based on the results from the case studies, we suggest that the most important is data governance.

Table 4.

Artificial Intelligence Antecedents.

AI Antecedents	SATIKAS	Child and Family Predictive System	AmberScript
High-quality data	High-quality data to develop model and get useable results	High-quality data to develop model and get useable results	Over 1,000 hr of high-quality audio and transcription data to train the model
Maintenance of data	Work to ensure data quality from Copernicus	Work for data maintenance to ensure high-quality inputs and outputs	Continuous data maintenance to improve the system
Data sharing	Public sector data from different organizations shared with private partners	Different data from public organizations required	Audio data gained with permission from public organizations and used by a private organization
Data-driven services ecosystem	Different forms of data/capabilities available in the local environment	Different forms of data/capabilities available in the organization	Availability of public data/capabilities for analysis from private sector contributed to the development of AI system

Figure 2 presents the conceptual framework revised after our case study analysis. It shows that the role of data governance is crucial for the development and adoption of AI, as most of the early literature suggests. However, as mentioned by one of the respondents, the quantity of data may not be the most important issue, as too much data could actually lessen the quality of the AI by generating correlations that are not present in reality. In contrast, high-quality data are crucial for training AI, relying on strong data management processes which may take a considerable amount of time and effort. Therefore, constant maintenance of AI systems is required to ensure high levels of performance. This is considered vital for guaranteeing trust in AI systems and for the sustainable adoption of AI.

Figure 2.

Revised conceptual model of innovation with artificial intelligence based on de Vries et al. (2016).

For this reason, while sharing of data can be related to the antecedents of networks, it merits being seen as an antecedent in its own as it seems to greatly stimulate the development and adoption of AI in the case of governmental processes and public services. AI systems used within public administrations rely on interorganizational data sharing, and value can be gained by both public and private actors.

Another AI-specific antecedent to be considered is linked to the increased datafication of societal processes within the ecosystem, which would likely lead to more data becoming available for the development of AI systems, in turn increasing the interest in AI in government due to the processing support it offers. This is worth of analysis, as the contextual information around data, including metadata codification and related semantic interoperability efforts, may help organizations to be more transparent about the inputs to AI systems and how content is used, while harnessing existing digital infrastructures.

The results of our analysis based on case studies confirm findings of early literature on AI in the public sector, pointing in particular to the recognized need for gaining high-quality data, data analytics capabilities, and strong data governance (Harrison et al., 2019). Our research is aligned with the emerging perspective on how smart technologies are adopted in the public sector to achieve public value (Criado & Gil-Garcia, 2019) and start filling an important empirical gap in this stream of investigation.

As a matter of fact, compared with the existing literature on public sector innovation, the adoption of AI could be well understood and researched as a new form of ICT-enabled governance innovation in the public sector (Misuraca & Viscusi, 2015), as it is indeed influenced by the same factors that impact other forms of innovation in the public sector as described in (De Vries et al., 2016) but requires specific data-governance antecedents that are a crucial element for the use and adoption of AI in government and public services.

In line with recent literature, the results of the case studies suggest how the technical infrastructure underlying the data ecosystem is needed for AI to be used as a prerequisite for AI adoption. It can thus be argued that a mature level of digital government is required for AI to be deployed successfully and that without a functioning data ecosystem including Internet of Things (IoT) systems and digital services, AI is likely not to be adopted as there is simply no data available to train the models (Kankanhalli et al., 2019). However, the adoption of these technologies come with their own hurdles (Janssen et al., 2017), so that administrations interested in using AI face both barriers in adopting the technical digital and data infrastructure and the implementation and use of AI applications.

Another specific element of AI-enabled innovation compared to other innovations in the public sector is the increased risk of social, economic, political, and ethical challenges that may emerge following its adoption (Dwivedi et al., 2019). For instance, traditional cultural and power relationships within government or toward citizens may change with the adoption of AI, making some government actors hesitant to adopt AI solutions, while others may in fact push for adoption.

Moreover, the “black box” nature of AI-enabled decisions limits the accountability of decisions, already reducing opportunities for citizens and internal staff to challenge recommendations provided by the AI due to lack of time, repercussions from supervisors, or the perceived legitimacy of the AI (Kuziemski & Misuraca, 2020). While previous eGovernment research has highlighted that public managers may use ICTs to reinforce their position (Kraemer & King, 2006), with AI, the gap between citizens and public administrations may even increase if AI applications remain difficult to scrutinize and to explain (Kuziemski & Misuraca, 2020). How civil servants perceive the value and challenges of AI, including the changing role of government, may have a large influence on how AI and similar technologies get adopted (Guenduez et al., 2020; Sun & Medaglia, 2019).

Conclusion

While the literature on the adoption of public sector innovation is quite established, there are research gaps in the analysis of the adoption and sustainable use of AI-enabled innovation in government and public services. In fact, several use cases of AI have been canceled after an initial successful adoption due to political, legal, or other reasons (Misuraca & van Noordt, 2020). Therefore, the study of AI adoption should adopt a long-term lens in order to witness if AI applications are in fact still in use after a certain time—but also to better understand the consequences of its adoption (Bailey & Barley, 2019). Changes in law, organizational structure, data quality, citizen, or staff resistance may unexpectedly end AI adoptions but have not received sufficient research attention yet.

In this perspective, our research aims at shedding light on how the various antecedents of public sector innovation can lead to sustainable and long-term adoption of AI innovations in the government. In doing so, we focus on the specific elements that characterize AI-enabled innovation, and in particular the data governance and ecosystem underpinning its use and adoption.

In addition, the insights from the analysis of case studies provide concrete indications to policy makers aiming to stimulate and develop the use of AI within their administrations. Whereas a clear policy intention to support the development of AI-enabled services within the public sector is emerging in the EU, as highlighted in the recent White Paper on AI for instance, policy responses may be more successful if focusing on multiple innovation antecedents such as funding, public–private sector partnerships, and citizens perceptions, rather than on solely improving data quality and data quantity, as it is instead often the main concern in technical research on AI and technooptimistic policy interventions stated in strategic documents.

To this end, the holistic framework proposed in this article gives a comprehensive overview of the key factors contributing to the successful adoption of AI systems. At the same time, results from the analysis of the case studies offer valuable insights for practitioners and public managers who are interested in adopting AI in their organizations, going beyond the narrow focus on data, processing power and algorithm development often highlighted in the mainstream AI literature and policy discourse so far.

However, as the use of AI in the public sector is still in its infancy, more research is clearly needed to truly understand how systems are adopted by public sector organizations, helping to validate and potentially adjust the AI-specific innovation antecedents identified in this article. Due to its exploratory nature, the research underpinning this article has several limitations that we aim to address in further research. First, the way in which “AI” has been conceptualized may create challenges for the applicability of the findings for the broad set of AI technologies. Another limiting factor is that the research focuses on a limited set of cases with little information available on organizational and individual factors contributing to specific AI adoption. In future research, it is therefore important to look at the wider context of the adoption of any digital government project, requiring facts and opinion—gathering from several actors working with or within the administrations involved that may have shaped the adoption of AI systems in government. This would require also looking more specifically at the entire data ecosystem that nurtures the development of AI-enabled innovation in government and public services, and how different data governance regimes may stimulate development and facilitate use, while promoting cross-fertilizing mechanisms for AI adoption in the public sector.

Footnotes

Acknowledgments

Great gratitude goes to Anu Masso from TalTech for academic direction of the thesis and Alessandro Annoni for allowing the research to be conducted in collaboration with Joint Research Centre (JRC), under the scientific supervision of Gianluca Misuraca, at that time leading JRC research and policy support on AI for the public sector as part of the AI Watch, a joint initiative of the European Commission’s JRC in collaboration with DG CONNECT.

Authors’ Note

This article is based on research conducted for the final Thesis of Colin van Noordt as part of the PIONEER-Master Programme while visiting the European Commission’s Joint Research Centre in Seville, in January–June 2019. However, the views expressed in this article are purely those of the authors and may not be regarded as stating the official position of the European Commission.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Colin van Noordt is a PhD Researcher at the Ragnar Nurkse Department of Innovation and Governance at Tallinn University of Technology (TalTech), Estonia (coliva@ttu.ee).

Gianluca Misuraca is an independent strategic policy advisor on Digital Governance, Technology Diplomacy and Human-Centric Artificial Intelligence and a Research Fellow at the Department of eGovernance and Public Administration of the Danube University in Krems, Austria (gm@donau-uni.ac.at).

References

Agarwal

P. K.

(2018). Public administration challenges in the world of AI and Bots. Public Administration Review, 78(6), 917–921. https://doi.org/10.1111/puar.12979

Alexopoulos

Lachana

Androutsopoulou

Diamantopoulou

Charalabidis

Loutsaris

M. A.

(2019). How machine learning is changing e-government. In Proceedings of the 12th International Conference on Theory and Practice of Electronic Governance—ICEGOV2019, Part F1481 (pp. 354–363). https://doi.org/10.1145/3326365.3326412

Androutsopoulou

Karacapilidis

Loukis

Charalabidis

(2019). Transforming the communication between citizens and government through AI-guided chatbots. Government Information Quarterly, 36(2), 358–367. https://doi.org/10.1016/j.giq.2018.10.001

Bailey

D. E.

Barley

S. R.

(2019). Beyond design and use: How scholars should study intelligent technologies. Information and Organization, 30(2), 100286. https://doi.org/10.1016/j.infoandorg.2019.100286

Barocas

Selbst

A. D.

(2016). Big data’s disparate impact. California Law Review, 671(2016), 62.

Baxter

Jack

(2008). Qualitative case study methodology: Study design and implementation for novice researchers. Qualitative Case Study Methodology: Study Design and Implementation, 13(4), 544–559.

Bekkers

Tummers

Stuijfzand

B. G.

Voorberg

(2013). Social innovation in the public sector: An integrative framework (Lipse Working Paper Series No. 1), pp. 1–53. http://www.lipse.org/userfiles/uploads/Working%20paper%201%20Bekkers%20et%20al.pdf

Bertot

Estevez

Janowski

(2016). Universal and contextualized public services: Digital public service innovation framework. Government Information Quarterly, 33(2), 211–222. https://doi.org/10.1016/j.giq.2016.05.004

Bleive

(2017). Satellite based mowing detection. https://ec.europa.eu/jrc/sites/jrcsh/files/bleive.pdf

10.

Bleive

Voormansik

(2016). Satellite based grassland mowing detection. Estonian Agricultural Registers and Information Board.

11.

Bongers

De Bruyn

Verlet

(2018). Small kids, big data. Toepassing analytics in de kinderopvang in Vlaanderen. Vlaams Tijdschrift voor Overheidsmanagement (VTOM), 2018(1), 63–73. https://www.jurisquare.be/en/journal/vtom/2018-1/small-kids-big-data-toepassing-analytics-in-de-kinderopvang-in-vlaanderen/

12.

Borins

(2000). Loose cannons and rule breakers, or enterprising leaders? Some evidence about innovative public managers. Public Administration Review, 60(6), 498–507. https://doi.org/10.1111/0033-3352.00113

13.

Burrell

(2016, January). How the machine “Thinks:” Understanding opacity in machine learning algorithms. Big Data & Society, 1–12. https://doi.org/10.2139/ssrn.2660674

14.

Centre for Public Impact. (2017). Destination unknown: Exploring the impact of artificial intelligence on government. https://resources.centreforpublicimpact.org/production/2017/09/Destination-Unknown-AI-and-government.pdf

15.

Chui

Manyika

Miremadi

Henke

Chung

Nel

Malhotra

(2018). Notes from the AI frontier. Insights from hundreds of use cases. McKinsey Global Institute. https://www.mckinsey.com/∼/media/McKinsey/Featured Insights/Artificial Intelligence/Notes from the AIfrontier Applications and value of deep learning/Notes-from-the-AI-frontier-Insights-from-hundreds-of-use-cases-Discussion-paper.ashx

16.

Cinar

Trott

Simms

(2018). A systematic review of barriers to public sector innovation process. Public Management Review, 21(2), 1–27. https://doi.org/10.1080/14719037.2018.1473477

17.

Commission

(2017). How Copernicus is paving the way for the future of the CAP and Farming 2.0. https://www.copernicus.eu/en/how-copernicus-paving-way-future-cap-and-farming-20

18.

Company Webcast. (2019). Politieke taalmodel verhoogt kwaliteit (automatisch) ondertitelen [Political language model increases (automatic) subtitling quality]. https://www.companywebcast.com/nl/nieuws/politieke-taalmodel-verhoogt-kwaliteit-automatisch-ondertitelen/

19.

Craglia

Annoni

Benczur

Bertoldi

Delipetrev

De Prato

Feijoo

Fernandez-Macias

Gomez

Iglesias

Junklewitz

Lopez-Cobo

Martens

Nascimento

Nativi

Polvora

Sanchez

Tolan

Tuomi

… Vesnic Alujevic

(2018). Artificial Intelligence—A European perspective ( Craglia

, Ed.). Publications Office. https://doi.org/10.2760/11251

20.

Criado

J. I.

Gil-Garcia

J. R.

(2019). Creating public value through smart technologies and strategies: From digital services to artificial intelligence and beyond. International Journal of Public Sector Management, 32(5), 438–450. https://doi.org/10.1108/IJPSM-07-2019-0178

21.

De Vries

Bekkers

Tummers

(2016). Innovation in the public sector: A systematic review and future research agenda. Public Administration, 94(1), 146–166. https://doi.org/10.1111/padm.12209

22.

de Vries

Tummers

Bekkers

(2018). The diffusion and adoption of public sector innovations: A meta-synthesis of the literature. Perspectives on Public Management and Governance, 1(3), 159–176. https://doi.org/10.1093/ppmgov/gvy001

23.

Desouza

K. C.

Dawson

G. S.

Chenok

(2020). Designing, developing, and deploying artificial intelligence systems: Lessons from and for the public sector. Business Horizons, 63(2), 205–213. https://doi.org/10.1016/j.bushor.2019.11.004

24.

Dwivedi

Y. K.

Hughes

Ismagilova

Aarts

Coombs

Crick

Duan

Dwivedi

Edwards

Eirug

Galanos

Ilavarasan

P. V.

Janssen

Jones

Kar

A. K.

Kizgin

Kronemann

Lal

Lucini

… Williams

M. D.

(2019, August). Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management. https://doi.org/10.1016/j.ijinfomgt.2019.08.002

25.

Eggers

Schatsky

Viechnicki

Eggers

D. W.

(2017). AI-augmented government: Using cognitive technologies to redesign public sector work. Deloitte Center for Government Insights. https://www2.deloitte.com/content/dam/insights/us/articles/3832_AI-augmented-government/DUP_AI-augmented-government.pdf

26.

Engstrom

D. F.

D. E.

Sharkey

C. M.

Cuéllar

M.-F.

(2020). Government by algorithm: Artificial intelligence in federal administrative agencies. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3551505

27.

European Commission. (2018). About AI watch. https://ec.europa.eu/knowledge4policy/ai-watch/about_en

28.

European Commission. (2019, December 24). Organisation of childcare. https://eacea.ec.europa.eu/national-policies/eurydice/content/organisation-childcare-0_en

29.

Floridi

(2017). Group privacy: A defence and an interpretation. In Group privacy (pp. 83–100). Springer International Publishing. https://doi.org/10.1007/978-3-319-46608-8_5

30.

Flyvbjerg

(2006). Five misunderstandings about case-study research. Qualitative Inquiry, 12(2), 219–245. https://doi.org/10.1177/1077800405284363

31.

Guenduez

A. A.

Mettler

Schedler

(2020). Technological frames in public administration: What do public managers think of big data? Government Information Quarterly, 37(1), 101406. https://doi.org/10.1016/j.giq.2019.101406

32.

Harrison

Luna-Reyes

L. F.

Pardo

De Paula

Najafabadi

Palmer

(2019). The data firehose and AI in government. 171–176. https://doi.org/10.1145/3325112.3325245

33.

Hinings

Gegenhuber

Greenwood

(2018). Digital innovation and transformation: An institutional perspective. Information and Organization, 28(1), 52–61. https://doi.org/10.1016/j.infoandorg.2018.02.004

34.

Janssen

Konopnicki

Snowdon

J. L.

Ojo

(2017). Driving public sector innovation using big and open linked data (BOLD). Information Systems Frontiers, 19(2), 189–195. https://doi.org/10.1007/s10796-017-9746-2

35.

Jun

K. N.

Weare

(2011). Institutional motivations in the adoption of innovations: The case of e-government. Journal of Public Administration Research and Theory, 21(3), 495–519. https://doi.org/10.1093/jopart/muq020

36.

Kamal

M. M.

(2006). IT innovation adoption in the government sector: Identifying the critical success factors. Journal of Enterprise Information Management, 19(2), 192–222. https://doi.org/10.1108/17410390610645085

37.

Kankanhalli

Charalabidis

Mellouli

(2019). IoT and AI for smart government: A research agenda. Government Information Quarterly, 36(2), 304–309. https://doi.org/10.1016/j.giq.2019.02.003

38.

Katz

(2017). Manufacturing an artificial intelligence revolution. SSRN, 1–21. https://doi.org/10.2139/ssrn.3078224

39.

Kraemer

King

J. L.

(2006). Information technology and administrative reform: Will e-government be different? International Journal of Electronic Government Research (IJEGR), 2(1), 1–20. https://doi.org/10.4018/jegr.2006010101

40.

Krafft

P. M.

Young

Katell

Huang

Bugingo

(2019). Defining AI in policy versus practice. http://arxiv.org/abs/1912.11095

41.

Kuziemski

Misuraca

(2020). AI governance in the public sector: Three tales from the frontiers of automated decision-making in democratic settings. Telecommunications Policy, 44(6), 101976. https://doi.org/10.1016/j.telpol.2020.101976

42.

Lewis

J. M.

Ricard

L. M.

Klijn

E. H.

(2018). How innovation drivers, networking and leadership shape public sector innovation capacity. International Review of Administrative Sciences, 84(2), 288–307. https://doi.org/10.1177/0020852317694085

43.

Lima

M. S. M.

(2020). Predicting and explaining corruption across countries: A machine learning approach. Government Information Quarterly, 37(1). https://doi.org/10.1016/j.giq.2019.101407

44.

Makridakis

(2017). The forthcoming artificial intelligence (AI) revolution: Its impact on society and firms. In Futures (Vol. 90, pp. 46–60). https://doi.org/10.1016/j.futures.2017.03.006

45.

Mehr

(2017). Artificial intelligence for citizen services and government. https://ash.harvard.edu/files/ash/files/artificial_intelligence_for_citizen_services.pdf

46.

Meijer

(2015). E-governance innovation: Barriers and strategies. Government Information Quarterly, 32(2), 198–206. https://doi.org/10.1016/j.giq.2015.01.001

47.

Ministerie van Binnenlandse Zaken en Koninkrijksrelaties. (2017). Huidige beleid: Besluit digitale toegankelijkheid [Current policy: Decision on digital accessibility]. https://www.digitoegankelijk.nl/beleid/wet-en-regelgeving/huidig-beleid-besluit-digitale-toegankelijkheid

48.

Misuraca

van Noordt

(2020). AI Watch—Artificial Intelligence in public services. https://doi.org/10.2760/039619

49.

Misuraca

Viscusi

(2015). Shaping public sector innovation theory: An interpretative framework for ICT-enabled governance innovation. Electronic Commerce Research, 15(3), 303–322. https://doi.org/10.1007/s10660-015-9184-5

50.

Mittelstadt

B. D.

Allo

Taddeo

Wachter

Floridi

(2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2). https://doi.org/10.1177/2053951716679679

51.

Network of European Regions Using Space Technologies [NEREUS]. (2018). The ever growing use of COPERNICUS across Europe’s regions. European Commission.

52.

Organisation for Economic Co-operation and Development [OECD]. (2019). Digital opportunities for better agricultural policies. In Digital opportunities for better agricultural policies. OECD Publishing. https://doi.org/10.1787/571a0812-en

53.

Pasquale

(2015). The black box society. Harvard University Press. https://doi.org/10.4159/harvard.9780674736061

54.

Potts

Kastelle

(2010). Public sector innovation research: What’s next? Innovation: Management, Policy and Practice, 12(2), 122–137. https://doi.org/10.5172/impp.12.2.122

55.

Preece

Ashelford

Armstrong

Braines

(2018). Hows and whys of artificial intelligence for public sector decisions: Explanation and evaluation. http://arxiv.org/abs/1810.02689

56.

Rogers

E. M.

(2010). Diffusion of innovations (4th ed.). Simon & Schuster.

57.

Russel

S. J.

Norvig

. (2016). Artificial intelligence—A modern approach (4th ed.). Pearson.

58.

Schedler

Guenduez

A. A.

Frischknecht

(2019). How smart can government be? Exploring barriers to the adoption of smart government. Information Polity, 24(1), 3–20. https://doi.org/10.3233/IP-180095

59.

Scherer

M. U.

(2016). Regulating artificial intelligence systems: Risks, challenges, competencies, and strategies. Harvard Journal of Law & Technology, 29(2), 353–400.

60.

Sørensen

Torfing

(2011). Enhancing collaborative innovation in the public sector. Administration and Society, 43(8), 842–868. https://doi.org/10.1177/0095399711418768

61.

de Sousa

W. G.

de Melo

E. R. P.

Bermejo

P. H. D. S.

Farias

R. A. S.

Gomes

A. O.

(2019, July). How and where is artificial intelligence in the public sector going? A literature review and research agenda. Government Information Quarterly. https://doi.org/10.1016/j.giq.2019.07.004

62.

Stone

Brooks

Brynjolfsson

Calo

Etzioni

Hager

Hirschberg

Kalyanakrishnan

Kamar

Kraus

Leyton-Brown

Parkes

Press

Saxenian

Shah

Tambe

Teller

(2016).Artificial intelligence and life in 2030: One hundred year study on artificial intelligence. Stanford University, 52. https://doi.org/https://ai100.stanford.edu

63.

Sun

T. Q.

Medaglia

(2019). Mapping the challenges of artificial intelligence in the public sector: Evidence from public healthcare. Government Information Quarterly, 36(2), 368–383. https://doi.org/10.1016/j.giq.2018.09.008

64.

Susar

Aquaro

(2019, April). Artificial intelligence: Opportunities and challenges for the public sector. In Proceedings of the 12th International Conference on Theory and Practice of Electronic Governance (pp. 418–426). Association for Computing Machinery. https://doi.org/10.1145/3326365.3326420

65.

Tartu Observatory. (2019). Information system SATIKAS helps to detect mowing by using satellite data. https://kosmos.ut.ee/en/news/information-system-satikas-helps-detect-mowing-using-satellite-data

66.

van Noordt

Misuraca

(2019). New wine in old bottles: Chatbots in government. In Panagiotopoulos

(Ed.), Electronic participation. ePart 2019. Lecture Notes in Computer Science. Springer. https://doi.org/10.1007/978-3-030-27397-2_5

67.

Veale

Van Kleek

Binns

(2018, April). Fairness and accountability design needs for algorithmic support in high-stakes public sector decision-making. In Proceedings of Conference on Human Factors in Computing Systems (pp. 440:1–440:14). https://doi.org/10.1145/3173574.3174014

68.

Vinuesa

Azizpour

Leite

Balaam

Dignum

Domisch

Felländer

Langhans

S. D.

Tegmark

Fuso Nerini

(2020). The role of artificial intelligence in achieving the sustainable development goals. Nature Communications, 11(1), 233. https://doi.org/10.1038/s41467-019-14108-y

69.

Wirtz

B. W.

Weyerer

J. C.

Geyer

(2019). Artificial 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

70.

Yin

R. K.

(2018). Case study research: Design and methods (6th ed.). Sage Publications.