Experimenting with an Artificial Intelligence Initiative in a Public Administration: Contributions from an Action Research Approach

Experimenting with an Artificial Intelligence Initiative in a Public Administration: Contributions from an Action Research Approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Experimenting with an Artificial Intelligence Initiative in a Public Administration: Contributions from an Action Research Approach Montassar Bahri, Mahmoud ZOUAOUI This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8290777/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Advances in Artificial Intelligence (AI) and its growing presence in professional settings raise questions about the transformation of public administrations, which are often addressed in speculative terms. This article seeks to demonstrate how a public administration can implement an AI initiative within its services. Adopting a constructivist stance and an action research approach (Susman & Evered, 1978 ), the study draws on iterative cycles between theory and field inquiry. Based on semi-structured interviews with relevant stakeholders and participant observation, the article relies on three successive qualitative studies conducted in three organizational units of the Tunisian Ministry of the Interior. The findings highlight the need to anticipate the impacts of AI tools by evaluating them in their actual context of use while actively involving the stakeholders concerned. They underscore the importance of design processes grounded in genuine needs identified by members of the organization themselves. Finally, the results identify the key success factors and the challenges associated with AI integration, showing that such implementation can strengthen dynamic capabilities and knowledge management, thereby fostering a more effective adoption of technologies within public administrations. Artificial Intelligence Public Administration Dynamic Capability Knowledge Management Sentiment Analysis Figures Figure 1 Figure 2 Figure 3 Introduction In the context of transformations driven by the knowledge economy, managers are required to anticipate changes in their environment (Nutt & Backoff, 1997 ). AI has emerged as a major lever of this evolution, reshaping academic, economic, and geopolitical spheres (Parly, 2019 ). Supported by the abundance of data, increased computational power, and the emergence of tools such as Matplotlib and NumPy (Anderson, 2014), AI now enables the forecasting of environmental developments and the automation of tasks requiring intelligence, speed, and precision. Initially developed in the private sector, AI now attracts growing interest within public administrative organizations, where it opens new avenues for action. However, the literature emphasizes the need to consider potential risks and negative effects (Wirtz, 2019) and to favor simple, well-structured pilot projects. Few studies have simultaneously integrated the technical, human, and social dimensions involved in the implementation of an AI initiative. Moreover, research remains limited regarding the value of AI in security institutions, even though it can potentially enhance their operational performance (Valérie D., 2019 ). Scholars such as Mazzucchi and Guyonneau (2019) support this view, though without proposing a concrete implementation framework. While AI applications in the security field are not new, their renewed relevance lies in the emergence of tools that enable more effective information exploitation. For the Tunisian Ministry of the Interior, the most relevant applications relate to data analysis and the detection of criminal threats. This study therefore seeks to design an AI initiative tailored to the Ministry’s analysis and research units, with the aim of improving the speed, accuracy, and reliability of their missions. Its objective is to identify the key success factors for such an initiative, propose appropriate methodologies, and support staff in their appropriation processes. Beyond the technological dimension, implementation requires an evolution of organizational values and behaviors (Santiso, 2019). An evaluation of the results will help assess both technical and behavioral changes, identify the challenges encountered, and formulate ethical recommendations that reconcile administrative values with AI-driven improvements. Based on the foregoing, the research question guiding this study is as follows: How can a public administration implement an effective AI initiative within its information analysis and research services? This research adopts a constructivist stance within the action research paradigm. It mobilizes theory and practice to develop a conceptual framework and design an AI tool appropriate to the administrative context. The selection of the Tunisian Ministry of the Interiorgiven the strategic role of the security sectorenables assessment of the relevance of such an initiative and the proposal of a transferable integration strategy for other public administrations, through a qualitative approach grounded in case studies and data triangulation. The article presents a study structured into three phases, guided by Susman and Evered’s ( 1978 ) model. The first phase involves diagnosing the perspectives of actors within the Tunisian Ministry of the Interior through ten interviews and participant observation, with the aim of identifying needs and challenges related to the implementation of an AI initiative. The second phase, based on an iterative learning process linking theory and practice, focuses on designing and launching the initiative in collaboration with the relevant actors, supported by ten additional interviews. The third phase evaluates the implementation in order to identify its benefits and the difficulties encountered at the material, organizational, and human levels, drawing on ten further interviews conducted with the same participants. Conceptual Framework Defining AI remains a complex task, given the breadth and interdisciplinary nature of its field of application. Limiting it to a single area of research would be reductive, as its foundations draw simultaneously from computer science, statistics, cognitive sciences, management, and organizational theory. From this perspective, it is relevant to mobilize several key concepts, particularly dynamic capabilities, knowledge management, and AI itself,in order to better understand the interactions that connect them. A joint analysis of these notions helps clarify the scope of the AI concept and identify the theoretical foundations that can inform its integration within organizations. An Integrative Reading of Dynamic Capabilities in Light of Artificial Intelligence Technologies The Resource-Based View (RBV), introduced by Wernerfelt ( 1984 ), explains organizational performance through the nature and characteristics of the resources held by the firm. According to Barney ( 1991 ), resources can generate a competitive advantage when they are valuable and rare, and a sustainable advantage when they are also inimitable and non-transferable. Such resources enable organizations to seize opportunities, counter external threats, andthrough causal ambiguitymaintain an advantage that remains difficult for competitors to identify. However, several authors (Priem & Butler, 2001a , 2001b) have criticized the RBV for its static conception of performance, which is poorly suited to unstable environments. In response, the dynamic capabilities approach (Teece et al., 1997 ; Wang & Ahmed, 2007 ) emerged as an evolutionary extension of the RBV, emphasizing the ability of organizations to adapt, integrate, and reconfigure their resources in the face of change. Subsequent empirical work (Yami et al., 2008 ) further developed this perspective by identifying the factors that support the development of dynamic capabilities and their contribution to organizational competitiveness. Wang and Ahmed ( 2007 ) view resources as the foundation of any organization, while distinguishing them from capabilities. In this hierarchy, functional capabilities occupy the next level, corresponding to the organization’s ability to perform fundamental activities such as production, marketing, or distribution (Collis, 1994 ). At the following level are core capabilities, which encompass the resources, skills, and knowledge that provide a sustainable competitive advantage (Leonard-Barton, 1992; Collis, 1994 ; Wang & Ahmed, 2007 ). Finally, dynamic capabilities stand at the top of this hierarchy, as they reflect an organization’s ability to transform and reconfigure its resources in response to change (Winter, 2003). These capabilities extend beyond mere adaptation and are rooted in structured organizational learning. Unlike ad hoc problem-solvingwhich relies on improvisationdynamic capabilities result from a continuous process of learning and preparation for change. As Josserand ( 2007 ) argues, the dynamic recombination of resources is itself the outcome of an ongoing learning dynamic. A Theoretical Perspective on Artificial Intelligence as a Support for Knowledge Creation and Sharing AI contributes significantly to organizational knowledge management by facilitating the creation, storage, sharing, and application of knowledge. It enables the transformation of individual knowledge into collective and organizational knowledge, thereby enhancing organizations’ capacity for learning and adaptation. Knowledge Creation AI, particularly deep learning, facilitates the discovery of new knowledge by detecting hidden patterns within data. It can predict trends (e.g., sales) and generate new insights from existing datasets. Practical applications, such as the analysis of millions of scientific publications, demonstrate that AI can identify previously unrecognized correlations, opening new avenues for research and innovation. In this way, AI helps organizations leverage big data and extract value from information that is often underutilized (with some studies suggesting that over 60% of data remains unused). Knowledge Storage and Retrieval AI enhances organizational memory by enabling the classification, organization, and retrieval of large volumes of data. Machine learning algorithms allow knowledge to be extracted and structured from multiple sources, as exemplified by recommendation systems (e.g., Gmail suggesting recipients). Consequently, AI makes knowledge bases more accessible, dynamic, and actionable. Knowledge Sharing Intelligent systems connect employees working on similar issues and foster collaborative intelligence. Platforms such as MITRE or collaborative tools like Slack and Yammer utilize AI to link individuals, stimulate collective creativity, and create a shared memory within teams. Knowledge Application AI facilitates the practical application of knowledge by tailoring it to specific needs and automating certain tasks. Intelligent assistants make relevant knowledge available in real time, thereby improving operational performance. For instance, Repsol has employed AI to drastically reduce unproductive time on its drilling sites through better data utilization. In customer service, platforms like Talla automate knowledge management to respond more rapidly to emerging needs. Artificial Intelligence Based on the various definitions of AI available in the literature, we have selected several that may help clarify how leaders perceive this concept. These definitions are summarized in the following table (Table 1 ): Table 1 Definitions of the Concept of AI Marvin Lee Minsky (1956) The development of computer programs that perform tasks currently accomplished more effectively by humans, as they require high-level mental processes such as perceptual learning, memory organization, and critical reasoning. A technology capable of producing results similar to those generated by the human brain. Harry Shum (2018) Operates only when there is a “vast amount of data; extraordinary computing power, particularly through the cloud; and groundbreaking algorithms based on deep learning.” McCarthy et al. (2006) The learning process or any other feature of artificial intelligence can, in principle, be described so precisely that a machine can be made to simulate it. Rich et al. (2019) It is the way computers perform tasks that, at present, humans perform better. Russel and Norvig, (2010) AI can be organized into four systems: systems that think like humans, systems that act like humans, systems that think rationally, and systems that act rationally. Rosa et al. (2016) Software capable of learning, adapting, being creative, and solving problems. Thierer et al. (2017) An AI system is one that can undertake high-level operations; AI can function at or beyond human capabilities. This concept is divided into weak AI and strong AI. Within the scope of this research, we adopt an operational definition of AI that is oriented toward the organizational and administrative context. : This definition primarily draws on the contributions of Russell and Norvig (2010), who distinguish between cognitive and rational approaches, as well as on McCarthy et al. (2006) regarding the simulation of human reasoning, while also integrating Harry Shum’s (2018) perspective on the importance of data and computational power as key enablers of contemporary artificial intelligence. It thus links the technological dimension of AI to an organizational purpose, consistent with the objectives of our action research focusing on the integration of an AI initiative within a public administration. Based on this operational definition, AI should now be considered from the perspective of its integration into organizational contexts. Beyond its technical and algorithmic aspects, AI emerges as a strategic instrument capable of supporting transformations in management practices, decision-making, and value creation within organizations. This conception extends beyond simple task automation, engaging a deeper reconfiguration of organizational processes, human interactions, and data governance. From this standpoint, AI appears not only as an emerging technology but also as a driver of organizational change that fosters collective learning and administrative innovation. Within this framework, the introduction of AI into public administrations raises a range of managerial challenges. Public organizations, historically shaped by hierarchical structures and a culture of compliance, are now confronted with the need to adapt their structures and practices to fully leverage AI’s potential. This involves not only exploiting the operational benefits of the technology but also anticipating its impacts on public decision-making, competency management, transparency, and citizen trust. It is these multifaceted challenges that we address in the following section.. Challenges in Implementing an AI Initiative within a Public Administration: Managerial Opportunities for Security Institutions Applications of AI in Public Administrative Organizations AI has generated considerable enthusiasm in the private sector in recent years, whereas the public sector still lags behind in AI adoption and deployment. Nevertheless, advances in AI offer a wide range of opportunities for its use within public organizations. Several research studies have examined AI applications and challenges only in a fragmented and isolated manner. Scholars have synthesized scientific publications on AI to provide a comprehensive overview of its use in organizations. While public administrative organizations are increasingly experimenting with AI applications, their adoption remains less widespread than in the private sector. An analysis of multiple publications on this topic has identified ten major categories of AI applications within public organizations (Wirtz et al., 2018). These applications include AI-based knowledge management software, process automation systems, virtual agents, as well as predictive analytics and data visualization tools. They also encompass identity analysis systems, cognitive robotics and autonomous systems, recommendation engines, intelligent digital assistants, discourse analysis tools, and cognitive security and threat detection systems. Collectively, these technologies illustrate the diversity of possible AI uses in the administrative sphere and its potential to transform the operational modes of public institutions. Benefits of AI Applications in Public Administrative Organizations AI not only drives technological innovation but also influences sociological, environmental, political, and administrative dimensions. Public organizations, as key actors in state development, are already witnessing AI integration across multiple areas of the public sector (Misuraca & Van Noordt, 2020 ; Androutsopoulou et al., 2019 ; Ojo et al., 2019 ; Sun & Medaglia, 2019 ). Although the pace and nature of this integration vary across countries, AI adoption is becoming progressively unavoidable in many government functions (De Sousa et al., 2019 ). Moreover, the rapid evolution of AI technologies opens new opportunities for designing and implementing public policies (Önder & Saygili, 2018). Today, AI usage extends to various public services, as evidenced by Egger et al. ( 2019 ), who document multiple examples and adoption rates within the sector. Several studies (Purdy & Daugherty, 2016 ; Gartner & Hiebl, 2017) highlight the benefits of AI in public administrative organizations, including improved time management, development of self-learning and co-learning capabilities, and broader dissemination of innovation. They also emphasize the effectiveness of AI applications in reducing costs and time, enhancing accessibility, and promoting the inclusion of public services. Empirical evidence confirms a marked trend toward AI utilization in security, surveillance, and internal process optimization. In the security domain, Valérie D. ( 2019 ) identifies several advantages of AI applications. These technologies enable more effective automatic detection of stealth targets and facilitate the exploitation of increasingly voluminous satellite data. They also provide valuable support to decision-makers by assisting them in making decisions within complex environments while maintaining operational speed. Furthermore, AI contributes to predictive maintenance, optimizing the availability of weapon systems, and plays a crucial role in detecting cyberattacks. It is also applied through robotic support on the battlefield or in aerial combat, as well as via human–machine interaction functions in natural language. Additionally, AI enhances the performance of telecommunication networks through agile and automated routing and strengthens the realism of military training systems. Many of these applications also have relevant civilian uses, demonstrating the transversal and strategic significance of AI in the field of security. Preparations for Implementing an AI Tool: Practical Framework (Data Collection No. 1) The diagnostic phase is based on semi-structured interviews aimed at capturing the perspectives of leaders within the Tunisian Ministry of the Interior to identify obstacles, benefits, and challenges associated with implementing an AI initiative, while also collecting their proposals. The interview guide is designed to understand the perceived barriers, opportunities, and advantages of this initiative, thereby contributing to the refinement of the research conceptual framework. In total, ten ministry officials were interviewed in this context. The action research model adopted is that of Susman and Evered ( 1978 ), recognized as one of the most widely used frameworks for describing this type of initiative (Baskerville, 1999). This model comprises five stages: diagnosis, action planning, action execution, evaluation, and knowledge generation, which are here applied to the implementation of an AI initiative within a public administration. The first two stagesdiagnosis and planningrequire continuous iteration between theory and fieldwork to verify and enrich the knowledge base through primary data collection with leaders of the Tunisian Ministry of the Interior. This data collection serves to refine the theoretical framework that will underpin the subsequent stages of AI implementation. Initial Field Exploration: Refining Challenges and Preparing the Implementation Phase This preliminary field exploration aims to determine the specific needs of the organizational context and guide the selection of the AI tool to be implemented within the public administration. It is positioned at this stage to refine understanding of the challenges and optimally prepare for the implementation phase. Analysis of the interviews reveals that leaders still rely heavily on traditional data analysis methods, while showing growing interest in innovative analytical tools. This openness reflects an explicit willingness to gradually introduce an AI initiative within their organizations, with the goal of improving decision-making practices and overall service performance. Interviewed leaders clearly perceive the benefits associated with implementing an AI initiative. They particularly highlight improvements in processing quality and speed, as well as enhanced capabilities for data collection and analysis, leading to more relevant and timely outcomes. Beyond the technological dimension, these benefits are also linked to the development of internal competencies and improved knowledge management, positioning AI as a lever for organizational transformation and collective learning. However, several barriers remain. The primary obstacles identified relate to the human dimension: resistance to change, perceived complexity, and lack of training. Staff may indeed struggle to adapt to new practices induced by AI. Therefore, adequate preparation is essential prior to any effective implementation. In this context, the role of leadership is central: leaders must ensure the technical readiness of the project, select and train staff capable of adopting these tools, and motivate them to support successful and sustainable AI adoption. According to the interviewed leaders, three main approaches can be distinguished regarding the design and integration of AI systems. The first involves the development of generic AI algorithms by private, national, or international companies specializing in creating tools or platforms not intended for a specific application. The second corresponds to the implementation of ready-to-use AI systems, relying on existing software whose functionalities can be leveraged for certain specific tasks. Finally, the third approachadopted in this researchconsists of developing an AI system based on a concrete need and a specific problem, in this case a sentiment analysis tool. Accordingly, the phases of implementing an AI initiative can be grouped according to these three approaches, with their frequency of occurrence depending on organizational specificities, as illustrated in the following table (Table 2 ) : Table 2 Different Phases of Design and Deployment Approaches According to the Officials Generic AI Algorithms Ready-to-Use AI System Case-Specific AI System 1. Investigation and Selection of the Company 2. Contact and Contract 3. AI Project Formulation: Requirements Gathering 4. AI System Development 5. Team Selection 6. Training 7. Tool Testing with Real Cases 8. Iteration and Evaluation 1. AI Project Formulation: Requirements Gathering 2. Ensuring Confidentiality (IT Security) 3. Selection of Tools Adaptable to the Service 4. Team Selection 5. Training 6. Tool Testing with Real Cases 7. Iteration and Evaluation 1. AI Project Formulation: Requirements Gathering 2. Selection of Tools Adaptable to the Use Case 3. Team Selectio 4. Specialized Training 5. Tool Testing with Real Cases 6. Iteration and Evaluation At the conclusion of this phase, the interviewees opted for the implementation of a sentiment analysis tool, considering that it would enable a better understanding of employees’ perceptions and opinions, identify friction points within internal processes, and support data-driven decision-making based on reliable qualitative insights. Sentiment Analysis The literature employs various terms to refer to sentiment analysis (SA), such as opinion mining, emotion analysis, subjectivity analysis, or evaluation of phenomena. According to Medhat et al. (2014), the most commonly used terms are opinion mining and sentiment analysis, which are considered equivalent and fall under the subfield of subjectivity analysis. These authors define sentiment analysis as the computational processing of opinions, emotions, and judgments expressed in a text, situated within the broader field of text mining research. However, some studies conflate the notions of sentiment and opinion: according to the Merriam-Webster dictionary, an opinion is a judgment or evaluation on a specific subject, whereas sentiment refers to an attitude or judgment influenced by emotion. Furthermore, the rise of social mediacharacterized by ease of communication and idea-sharing within communities (Kim, Sin & YooLee, 2014)has made these platforms among the most dynamic and influential information systems today (Lin, Yang & Li, 2020). The psychological or cognitive effects of social media data on user behavior cannot be fully measured (Crawford, 2009 ). Opinions and ideas shared on these platforms are varied and may sometimes be biased or manipulative (Jansen, Sobel & Cook, 2010 ). While widely shared posts can serve as valuable sources of information, their interpretation remains complex due to the diversity of emotions expressed, including negation, sarcasm, sadness, joy, anger, surprise, or concern (Ji et al., 2015 ; Ali et al., 2017 ; Zarrad et al., 2014 ). Nevertheless, exploring these emotions constitutes a strategic tool for understanding public dynamics and guiding political or economic decision-making (Chung, He & Zeng, 2015 ). Opinion mining and sentiment analysis in the context of big data have become increasingly useful, as they contribute to understanding human emotions by enabling the tracking of behavioral patterns when users engage with social media applications (Ji et al., 2016 ). Sentiment Analysis Process According to Alamoodi et al. ( 2021 ), the sentiment analysis process consists of four main stages (Fig. 1 ). It begins with source selection, which involves determining the platform from which sentiments will be extracted, such as Facebook, Twitter, YouTube, Instagram, or Threads (Chung et al., 2015 ). The next stage is data collection, during which information is gathered using hashtags (#) (Gayo-Avello et al., 2013) or keywords (Deng, Tang & Huang, 2015 ), in various forms such as posts, comments, news articles, or texts (Chung et al., 2015 ). The third stage is preprocessing, where collected data are cleaned and normalized to remove irrelevant elements such as repeated letters, stop words, or typographical errors, while extracting the necessary linguistic features (Baker et al., 2020 ; Pollacci et al., 2017 ). Finally, the sentiment analysis stage utilizes the preprocessed data, either automatically or manually, to identify the emotions expressed and their frequency (Deng et al., 2015 ). This structured process ensures a more precise and reliable understanding of opinions conveyed through digital content. According to Andrea Esuli and Fabrizio Sebastiani (2006), sentiment analysis also referred to as opinion mining has a wide range of applications. It enables tracking public attitudes toward a political party based on data from social media, electronic newspapers, or blogs. Similarly, it can be used to assess consumer perceptions of a product by automatically classifying online reviews. Rooted in natural language processing and text analysis, this technique aims to explore and categorize opinions in an automated manner. Its primary applications span multiple domains. In decision-making, opinions gathered from social media provide valuable behavioral insights that inform strategic organizational choices. In public policy development, online citizen opinion analysis can guide policymakers in designing policies better aligned with societal expectations. It is also particularly useful for detecting hate speech, facilitating the automated monitoring of offensive, provocative, or harmful content on blogs, news sites, and social platforms. Finally, in marketing research, this approach contributes to the study of products, services, and consumption trends, while also enabling the analysis of public attitudes toward government policies. Action Research as an Operational Framework We adopted the action research model of Susman and Evered ( 1978 ), widely applied in management sciences (Baskerville, 1999). This model comprises five stages: diagnosis, planning, action, evaluation, and learning. It provides a relevant framework for studying AI implementation as it combines knowledge generation with organizational transformation. The method promotes collaboration between researchers and practitioners, while allowing continuous adaptation of actions to the specificities of the administrative context. The research employed multiple intervention cycles, combining semi-structured interviews and participant observation with officials from the Tunisian Ministry of the Interior. These iterations allowed refinement of the conceptual framework and the real-world testing of a sentiment analysis AI tool as a decision-support mechanism and a means of modernizing public services. Alignment of Paradigm, Method, and Purpose The articulation between the constructivist paradigm and the action research approach confers both epistemological and methodological coherence to this study. Constructivism enables a situated and shared understanding of change, considering organizational reality as the product of interactions and interpretations of actors. The abductive approach, in turn, facilitates continuous interplay between theory and practice, allowing knowledge to be built from experience while being enriched by relevant conceptual frameworks. Finally, action research fosters a reflective dynamic in which knowledge informs action, and action, in turn, generates new knowledge. This alignment between paradigm, method, and purpose enhances the scientific and operational relevance of the adopted approach. Consequently, this study does not merely aim to describe a phenomenon but seeks to produce useful, transferable knowledge applicable to other public organizations. By integrating the researcher’s reflective stance and the involvement of field actors, the proposed approach contributes to strengthening organizational learning capacity and understanding the conditions for successful AI integration in the public sector. Methodological Choices A qualitative approach was deemed appropriate to analyze AI implementation within a security-focused public administration. This choice is justified by the nature of the phenomenon, which involves human, organizational, and managerial dimensions that are difficult to quantify. Consistent with Collerette (1996) and Hlady-Rispal (2002), the qualitative approach aims to understand the complexity of an evolving social system and to capture actors’ perceptions, barriers, and levers. Data collection relied on semi-structured interviews complemented by direct observation of organizational dynamics. Three interview guides were developed based on the literature review to ensure coherence between research objectives and the information collected. The first guide focused on the initial diagnosis of AI implementation, exploring perceived barriers, benefits, and challenges. The second guide addressed the implementation of a sentiment analysis tool, capturing technical and organizational integration aspects. Finally, the third guide aimed to evaluate the outcomes and capitalize on knowledge gained, producing lessons transferable to future similar initiatives. These guides were pre-validated by participants to ensure trust and improve data quality. Field Context The study was conducted within the Directorate General of Special Services of the Tunisian Ministry of the Interior, where three sub-directorates were selected due to their strong commitment to digital transformation. Sub-directorate A, focused on human development and organizational agility, leverages digital transformation to enhance analysts’ practices and efficiency. Sub-directorate B, despite possessing strong technical competencies, faces certain administrative constraints and a lack of analytical tools, limiting its innovation potential. Sub-directorate C, primarily oriented toward social media data collection and analysis, seeks to improve the reliability and performance of its analyses through AI integration. This organizational diversity allowed observation of different AI adoption dynamics according to the structural and cultural contexts of each unit, providing representative insight into the challenges faced in the Tunisian administration. Data Collection and Processing Data were collected between 2019 and 2024 within an action research framework. The researcher, embedded within the studied organization, conducted ten exploratory interviews followed by twenty semi-structured interviews during the implementation and evaluation phases. Interviews, averaging two hours each, involved leaders, administrative and technical managers, and the central director. In total, over 42 hours of formal interviews and several observation sessions were conducted. Interview settings were arranged to promote spontaneity and trust. The researcher’s role, oscillating between consultant and observer, minimized bias while fostering contextualized understanding. Probing questions (“Could you clarify?”, “Could you confirm?”) facilitated validation and coherence of the collected narratives. Data Analysis Data were transcribed, coded, and analyzed using MaxQDA, following an iterative coding process. An initial coding grid was developed in Excel and refined after analyzing sub-directorates A and B to better capture key dimensions of organizational change, technological barriers, and human levers. The final grid, applied to sub-directorate C and the central director, yielded structured thematic categories: success conditions for AI initiatives, human and organizational factors, leadership roles, and impacts on managerial practices. Triangulation of sources (interviews, observations, internal documents) strengthened the validity of the results and enabled an integrated reading of transformation dynamics. The analysis underscores the importance of human capital, training, and change management support for the successful implementation of AI initiatives in the public sector. Results We have chosen to present the results of the first phase of our research, corresponding to the diagnostic and action planning stages, alongside the exploration and enrichment of the theoretical framework. This preliminary phase allowed for the identification and supplementation of certain missing concepts. In particular, the analysis of exploratory interviews highlighted the necessity of integrating the concept of sentiment analysis, which was mentioned by the managers during the very first interview. This methodological choice aims to enhance the coherence between the conceptual framework and the empirical research process. The following section then presents the results of the second phase, concerning the implementation of the tool, as well as those of the third phase, focusing on the evaluation of the action. Results of the Second Phase: Implementation of a Sentiment Analysis Tool within the Tunisian Ministry of the Interior Before initiating the implementation of an AI tool across three sub-directorates, we sought to anticipate the expectations and potential changes within these units. The primary objective was to understand the perceived utility of the sentiment analysis tool and to identify doubts regarding its actual contributions, thereby aligning its potential with the operational and strategic needs of the examined entities. The deployment of the sentiment analysis tool primarily aimed to optimize analysts’ and operational actors’ working time, while improving the overall relevance of organizational activities. Initially, the project had no formal status or dedicated teams, as it originated from research initiatives and preliminary field interviews. The choice of the R programming language as a technical support tool was driven by several decisive advantages. R allows for results comparable to those obtained with paid software while integrating both backend (programming) and frontend (interface) functionalities within a single environment. It also enables all types of analyses to be performed freely using customizable scripts, without any licensing constraints. Therefore, adopting R met a dual requirement of efficiency and resource optimization, while fostering gradual appropriation of AI within a transforming administrative context. To address the lack of technical skills among interviewees, a detailed guide was developed explaining the research objectives, implementation steps, and technical aspects of using the tool. This guide was distributed to the central and general directorates of the Tunisian Ministry of the Interior with the approval of the Director General of National Security. The primary objective of the R tool was to simplify sentiment analysis, thereby optimizing analysts’ work and addressing the challenges posed by the absence of suitable tools for processing and interpreting textual data in a security context. As deployment progressed, the range of actors involved in the project diversified, and their roles gradually became defined. The main actors involved in the process and their respective contributions are illustrated in Fig. 3 . The figure above illustrates that the directors of the sub-directorates played a central role in identifying and formalizing organizational needs, while coordinating teams and maintaining liaison with hierarchical authorities. Specific managers served as points of contact between the directors and the analytical units, ensuring the transmission of directives and supervising the various stages of data collection. Analysts, as the primary users of the tool, actively participated in both the collection and analysis of sentiments, as well as in the practical exploitation of the results. Operational staff collaborated with analysts to guide data selection and support the tool’s testing phases. Finally, the researcher provided technical and methodological support by organizing specialized training on the R programming language, installing the tool on analysts’ workstations, and guiding them step by step in its use. This distribution of roles ensured a collaborative and gradual implementation of the sentiment analysis approach. All actors played dual roles, both in the design and deployment of the tool and as end-users, contributing simultaneously to its improvement and to the enhancement of the teams’ analytical capabilities. The introduction of the AI-based tool raised several questions regarding its actual utility and added value for the activities of the concerned units. AI tools, by enabling the processing of large volumes of data and synthesizing information, offer significant potential to facilitate and inform decision-making. Early feedback, however, highlighted several tangible benefits. The tool improved analysts’ performance and work quality, while contributing to the enhancement of human capacities by helping users better execute their daily tasks. Moreover, it promoted a more efficient exploitation of textual data, transforming it into relevant and actionable knowledge, particularly valuable in the security domain. The integration of this AI solution not only optimized analytical processes but also strengthened collective capacity to generate more nuanced and responsive decision intelligence. Thus, the implementation phase highlighted not only the anticipated benefits of the tool but also the conditions necessary for its successful adoption and integration within the examined sub-directorates. Following the implementation of the R-based sentiment analysis tool, several key success factors were identified to ensure the effective deployment of an AI initiative within a public administration. The process was structured around several interdependent steps. The first step consisted of collecting needs and data to clearly define organizational expectations and align them with the practical usefulness of AI tools. The second step involved selecting the appropriate AI tool based on the identified needs and on criteria such as effectiveness, functionality, cost, and ease of use. The third step focused on establishing a testing team composed of motivated and competent members responsible for piloting the implementation and monitoring the system. Comprehensive training was then provided to enable the team to fully exploit the tool’s potential through hands-on exercises. Finally, a continuous iteration and evaluation phase ensured the progressive adaptation of the tool to the organization’s specific needs and to the diversity of its application areas, including social, economic, political, educational, and counter-terrorism domains. This gradual and reflective approach underpins the sustainability and relevance of the AI initiative undertaken. Technical Challenges and Learning Opportunities The technical challenges encountered, such as managing dependencies among R packages, although initially perceived as obstacles, actually served as learning levers. They fostered the development of proactive strategies, continuous improvement of analytical practices, and the transfer of skills within the team. These technical difficulties thus became opportunities to strengthen both the efficiency and mastery of the tools, directly contributing to the success of the AI strategy. Based on our observations, implementing an AI initiative within an administrative organization entails several challenges related to training data. Data access varies across domains (security, economy, etc.), which can create difficulties in obtaining a significant data volume, sometimes leading to non-transparent acquisition methods. Data design must account for quantity, completeness, and feature quality, while data utilization requires prior cleaning to ensure analytical reliability. Consequently, the AI system cannot learn or improve autonomously without human intervention. Another critical challenge is the network effect: the more relevant data the AI integrates, the more effective it becomes. Data collection and cleaning phases require significant effort, and the quality of analytical results directly depends on the number of users and the amount of data exploited. We also observed that qualitative data analysis requires a team of at least four people to ensure effective implementation. Observed limitations include a perception of “handcrafted” design, concerns about reliability and reproducibility, and a sometimes overly technology-centered approach rather than one guided by actual organizational needs. These findings underscore the importance of aligning design with specific contextual requirements, particularly in security-related domains. Results of the Third Phase: Evaluation and Interpretation – AI at the Heart of the Administrative Organization To evaluate the implementation of an AI initiative within three administrative sub-directorates, we employed thematic content analysis. This approach allowed us to explore qualitative data from interviews and observations to understand the effects of the R-based sentiment analysis tool on organizational practices and anticipated changes. Enhancement of Competence Management The deployment of the R tool enabled analysts to acquire new theoretical and practical knowledge, significantly improving previously heterogeneous analytical practices. The action-research process involved formalizing explicit procedures, ensuring alignment between actors’ expectations and the AI tool’s potential. Field evaluation revealed that directors began reconfiguring resources and competencies within the sub-directorates, activating dynamic capabilities to implement rapid changes. Key steps in competence management can be summarized in four main axes: identification of skills and talents to pinpoint human resources capable of contributing effectively to organizational objectives; discovery and support of talents to foster their development and valorization; exploitation of skills by mobilizing them optimally to meet strategic and operational needs; and feedback integration to continuously adjust and improve competence management, promoting ongoing learning and organizational development. These observations confirm the dynamic capabilities model (Labrouche 2014), integrating organizational learning, strategic adaptation, and the creation of distinctive resources to respond to technological and organizational changes. New Vision of Knowledge Management Content analysis showed that introducing the R tool fostered a true culture of knowledge production and management among analysts. Its use encouraged quality-oriented behavior and continuous learning, generating reliable and actionable insights for decision-making. The knowledge management cycle observed can be divided into several complementary steps: cataloging, i.e., exhaustive collection of existing knowledge, including available resources and analysts’ know-how; structuring, to organize data for easier exploitation; verification, to ensure accuracy through collaboration with directors and the researcher; dissemination, to make knowledge accessible to relevant actors; and continuous optimization, to enhance practices and results based on feedback and new data, thus reinforcing organizational learning dynamics. This process transformed analysts’ behavior, fostering continuous learning and the production of high-quality analytical knowledge. Organizational Utility of the Tool The R tool demonstrated significant organizational value within sub-directorates A, B, and C. Directors were able to reassess and adapt strategies, integrating an organizational culture based on innovation and responsiveness to administrative evolution. The tool also enabled the acquisition of new management capabilities, reducing errors and improving analytical output quality. Impact analysis revealed multiple benefits within the organization. The tool enhanced effectiveness in achieving objectives while ensuring efficiency through optimal resource utilization. It facilitated intelligent exploration, enabling discovery and investigation of new information, and optimal use of collected data. Furthermore, it contributed to the continuous development of analysts’ skills while ensuring the relevance and reliability of produced results. Additionally, it supported timely decision-making and the creation of clear, innovation-based strategies. In the security domain, the tool facilitated preventive policing, risk anticipation, and a deeper understanding of root causes, helping to prevent recurrence. These combined effects illustrate the strategic and operational value that AI can provide within a complex organizational context. Thus, the R tool not only enhanced individual analytical skills but also reinforced organizational culture, generating short-, medium-, and long-term benefits for the organization. Discussion of Results This article explored the transformations induced by the design, deployment, and challenges associated with implementing an AI initiative within an administrative organization, particularly in the security domain. The three successive empirical studies conducted across three sub-directorates highlighted issues related to AI integration, the use and impact of AI tools on the quality of analytical outcomes, and the practical design and deployment process of the R-Studio tool, tailored to the specific needs of the sub-directorates under study. The main objective in designing the R tool was to train and guide analysts to adapt to its capabilities, develop their skills, and produce accurate and relevant results. Participants were involved either in identifying and formalizing requirements or directly contributing to the tool’s technical development. The design phase also included deployment in operational contexts, continuous optimization, and process improvement based on feedback from directors and managers. Several challenges were encountered before, during, and after implementation. Prior to deployment, identifying organizational needs proved complex, as actors were not always familiar with AI, and the administrative organization had no prior experience. Administrative instability complicated the selection of project leaders, and bureaucratic behaviors slowed necessary authorizations. Additionally, no strategy existed to evaluate internal competencies or to raise analyst awareness of AI tools. During implementation, hardware constraints, such as IT system incompatibilities and performance limitations, hindered tool usage. Analysts faced difficulties executing code and managing regular updates, while data security and confidentiality required strict protocols. Post-implementation, the absence of a strategy for technical adjustments and continuous training limited tool optimization, and analysts struggled to correctly interpret some outputs. The impacts of AI introduction were multifaceted. The tool improved performance and work quality, supporting decision-making, particularly in the analysis of social phenomena. However, these benefits are not universal and depend heavily on context, activities, and usage conditions. AI can sometimes generate inefficiencies or increased workloads if its limitations are not understood, and there is a risk of overconfidence or unrealistic expectations regarding the tool. The study showed that AI deployment mobilizes two major organizational levers. Dynamic capabilities manifested through the identification of opportunities provided by AI, progressive appropriation via experimentation, integration of tools into administrative tasks, and adjustment of workflows to fully exploit AI potential. Knowledge management was essential for structuring, externalizing, and internalizing tacit knowledge, consolidating best practices and operational procedures within work routines. The use of action research as an integrative framework allowed observation of these dynamics in real time, adjustment of interventions, and documentation of how dynamic capabilities and knowledge management were mobilized within a specific security context. Finally, ensuring the sustainability of the R tool remains a challenge. Its utility varies across sub-directorates and usage contexts, and its real relevance for decision-making may be questioned due to the confidentiality of information processed. Administrative instability and leadership transitions are also significant challenges. These elements highlight that real organizational performance benefits are difficult to measure, emphasizing the need for ongoing monitoring and a clear strategy to ensure sustained adoption of the AI tool. Conclusion This article examined the design and implementation of an AI initiative within an administrative organization, aiming to address the central question: how can a research and analysis service deploy an effective AI initiative? The study adopted a constructivist approach and action research as an integrative framework, combining field diagnostics, the design and deployment of an AI-based sentiment analysis tool, and evaluation of its implementation. Key results identified success factors such as data collection focused on organizational needs, appropriate selection of the tool and pilot team, adequate training, experimentation with real cases, and regular iterations for evaluation and adjustment. These factors highlighted not only observable impacts on analytical quality and work methods but also subtler managerial effects related to theoretical concepts such as dynamic capabilities and knowledge management. Methodologically, the abductive approach allowed addressing initial questions while deeply exploring the implications of tool deployment. Integration of Susman & Evered’s ( 1978 ) action research model facilitated knowledge acquisition and capitalization by actors, creating an organizational learning process. The multiple-case study design, covering the phases of diagnosis, planning, action, evaluation, and knowledge specification, enabled an in-depth analysis of AI-induced transformations from design to final assessment. Empirical contributions indicate that AI integration in administrative organizations can generate significant structural and organizational changes, improve analytical practices, and raise awareness among actors about the continuously evolving field. Tool evaluation involved the central director, directors, and managers of the three sub-directorates, allowing both theoretical and practical insights. This approach highlighted essential elements for successful implementation: defining processes, anticipating challenges, and considering impacts on activities, roles, and organization. The study also provides original contributions by exploring AI adoption in the Tunisian public sector, a relatively under-researched area, particularly in developing country administrations. It shows how dynamic capabilities and knowledge management enable organizational actors to progressively appropriate AI technologies by adapting resources and processes to technological changes. Initial AI deployments within the Tunisian Ministry of the Interior revealed significant gains in data processing and decision-making, emphasizing the importance of adequate preparation, continuous training, leadership support, and administrative flexibility. Finally, the research acknowledges certain limitations, including the mainly declarative nature of the first phase based on semi-structured interviews and the qualitative scope limited to three organizational units, which restricts generalizability. Future research could expand the sample, integrate a quantitative approach, and replicate the study across different contexts and AI tools to identify similarities and divergences, improving the proposed analytical framework. This approach could guide the design and deployment of AI systems in other administrative organizations and enhance understanding of AI’s organizational and managerial impacts. Declarations Ethics approval statement This research did not require formal approval from an institutional ethics committee, as it did not involve medical experimentation or sensitive personal data. The study was conducted in accordance with internationally recognized ethical principles for social science research. Participation was voluntary, participants were informed of the objectives of the study, and informed consent was obtained. All data were anonymized, and confidentiality was strictly maintained. Clinical trial number not applicable. Funding: The authors received no financial support for the research, authorship, and/or publication of this article. Author Contribution Authorship Contribution StatementM.B. conducted this research as part of his doctoral thesis and wrote the main manuscript. M.Z. supervised the work, provided methodological guidance, and contributed to the revision and refinement of the manuscript. All authors reviewed and approved the final version of the manuscript. References Alamoodi AH, Zaidan BB, Zaidan AA, Albahri OS, Mohammed KI, Malik RQ, Almahdi EM, Chyad MA, Tareq Z, Albahri AS (2021) Sentiment analysis and its applications in fighting COVID-19 and infectious diseases: A systematic review. Expert Syst Appl 167:114155. https://www.sciencedirect.com/science/article/pii/S0957417420308988 Ali K, Dong H, Bouguettaya A, Erradi A, Hadjidj R (2017) Sentiment analysis as a service: A social media based sentiment analysis framework. 2017 IEEE international conference on web services (ICWS), 660–667. https://ieeexplore.ieee.org/abstract/document/8029820/ Anderson N, Potočnik K, Zhou J (2014) Innovation and Creativity in Organizations: A State-of-the-Science Review, Prospective Commentary, and Guiding Framework. J Manag 40(5):1297–1333. https://doi.org/10.1177/0149206314527128 Androutsopoulou A, Karacapilidis N, Loukis E, Charalabidis Y (2019) Transforming the communication between citizens and government through AI-guided chatbots. Government Inform Q 36(2):358–367. https://www.sciencedirect.com/science/article/pii/S0740624X17304008 Araujo T, Helberger N, Kruikemeier S, De Vreese CH (2020) AI we trust? Perceptions about automated decision-making by artificial intelligence. AI Soc 35(3):611–623. https://doi.org/10.1007/s00146-019-00931-w Baker QB, Shatnawi F, Rawashdeh S, Al-Smadi M, Jararweh Y (2020) Detecting epidemic diseases using sentiment analysis of arabic tweets. J Univers Comput Sci 26(1):50–70. https://www.academia.edu/download/72331153/pdf.pdf Barney J (1991) Firm Resources and Sustained Competitive Advantage. J Manag 17(1):99–120. https://doi.org/10.1177/014920639101700108 Baskerville RL, Wood-Harper AT (1996) A critical perspective on action research as a method for information systems research. J Inform Technol 11(3):235–246. https://doi.org/10.1080/026839696345289 Baskerville R, Smithson S (1995) Information technology and new organizational forms: Choosing chaos over panaceas. Eur J Inform Syst 4(2):66–73. https://doi.org/10.1057/ejis.1995.8 Bernard F (2004) Constructivisme et sciences de l’organisation. De l’alternative au pluralisme épistémologique «limité». Communication Langages 139(1):27–40. https://www.persee.fr/doc/colan_0336-1500_2004_num_139_1_3249 Charreire S, Huault I (2001) Cohérence épistémologique et recherche en management stratégique Chung W, He S, Zeng DD (2015) eMood: Modeling Emotion for Social Media Analytics on Ebola Disease Outbreak Research-in-Progress Collis DJ (1994) Research Note: How Valuable are Organizational Capabilities? Strateg Manag J 15(S1):143–152. https://doi.org/10.1002/smj.4250150910 Crawford K (2009) Following you: Disciplines of listening in social media. Continuum 23(4):525–535. https://doi.org/10.1080/10304310903003270 De Sousa WG, de Melo ERP, Bermejo PHDS, Farias RAS, Gomes AO (2019) How and where is artificial intelligence in the public sector going? A literature review and research agenda. Government Inform Q 36(4):101392 Deng L, Yu D (2014) Deep learning: Methods and applications. Foundations and trends® in signal processing, 7(3–4) 197–387. https://www.nowpublishers.com/article/Details/SIG-039 Deng Xlong, Tang YQ, Huang YH (2015) Opinion mining for emergency case risk analysis in spark based distributed system. Proceedings of the 1st ACM SIGSPATIAL International Workshop on the Use of GIS in Emergency Management, 1–8. https://doi.org/10.1145/2835596.2835608 Dwivedi YK, Rana NP, Jeyaraj A, Clement M, Williams MD (2019) Re-examining the Unified Theory of Acceptance and Use of Technology (UTAUT): Towards a Revised Theoretical Model. Inform Syst Front 21(3):719–734. https://doi.org/10.1007/s10796-017-9774-y Egger F, Benzing V, Conzelmann A, Schmidt M (2019) Boost your brain, while having a break ! The effects of long-term cognitively engaging physical activity breaks on children’s executive functions and academic achievement. PLoS ONE 14(3):e0212482. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0212482 Esuli A, Sebastiani F, Abbasi A (2010) Sentiment Quantification. IEEE Intell Syst 25(4):72–75 Jansen BJ, Sobel K, Cook G (2010) Gen X and Ys attitudes on using social media platforms for opinion sharing. CHI ’10 Extended Abstracts on Human Factors in Computing Systems. 3853–3858. https://doi.org/10.1145/1753846.1754068 Ji R, Cao D, Zhou Y, Chen F (2016) Survey of visual sentiment prediction for social media analysis. Front Comput Sci 10(4):602–611. https://doi.org/10.1007/s11704-016-5453-2 Ji X, Chun SA, Geller J (2016) Knowledge-Based Tweet Classification for Disease Sentiment Monitoring. In W. Pedrycz & S.-M. Chen (Éds.), Sentiment Analysis and Ontology Engineering: An Environment of Computational Intelligence (pp. 425–454). Springer International Publishing. https://doi.org/10.1007/978-3-319-30319-2_17 Ji X, Chun SA, Wei Z, Geller J (2015) Twitter sentiment classification for measuring public health concerns. Social Netw Anal Min 5(1):13. https://doi.org/10.1007/s13278-015-0253-5 Josserand E (2007) Le pilotage des réseaux. Fondements des capacités dynamiques de l’entreprise. Rev Française de Gest 170(1):95–102. https://doi.org/10.3166/rfg.170.95-102 Kim G-H, Trimi S, Chung J-H (2014) Big-data applications in the government sector. Commun ACM 57(3):78–85. https://doi.org/10.1145/2500873 Kim K-S, Sin S-CJ, Yoo-Lee EY (2014) Undergraduates’ use of social media as information sources. https://dr.ntu.edu.sg/handle/10356/79680 Le Moigne J-L (1994) Le Constructivisme: Tome 1: Des fondements, Tome 2: Des epistemologies. ESF Lin SY, Shanafelt TD, Asch SM (2018) Reimagining clinical documentation with artificial intelligence. Mayo Clinic Proceedings, 93(5), 563–565. https://www.mayoclinicproceedings.org/article/S0025-6196(18)30142-3/abstract Maciejewski M (2017) To do more, better, faster and more cheaply: Using big data in public administration. Int Rev Admin Sci 83(1suppl):120–135. https://doi.org/10.1177/0020852316640058 Mazzucchi N (2019) IA et industrie de défense, le grand défi. Revue Défense Nationale 5:86–92. https://www.cairn.info/revue-defense-nationale-2019-5-page-86.htm Mehr H (2017) Artificial Intelligence for Citizen Services and Government Misuraca G, Van Noordt C (2020) AI Watch-Artificial Intelligence in public services: Overview of the use and impact of AI in public services in the EU. JRC Research Reports, JRC120399. https://ideas.repec.org/p/ipt/iptwpa/jrc120399.html Nutt PC, Backoff RW (1997) Organizational Transformation. https://journals.sagepub.com/doi/abs/10.1177/105649269763009 Ojo A, Raman M, Downe A (2019) Title: Toward green computing practices: A Malaysian study of green belief and attitude among Information Technology professionals. J Clean Prod 224. https://doi.org/10.1016/j.jclepro.2019.03.237 Onder M, Saygili H (2018) YAPAY ZEK VE KAMU YÖNETİMİNE YANSIMALARI /. ARTIFICAL INTELLEGENCE AND THE REFLECTIONS ON PUBLIC ADMINISTRATION Parly F (2019) Avant-propos – IA et défense. Revue Défense Nationale 820(5):9–17. https://doi.org/10.3917/rdna.820.0009 Pencheva I, Esteve M, Mikhaylov S (2018) Big Data and AI – A transformational shift for government: So, what next for research? Public Policy Adm 35:095207671878053. https://doi.org/10.1177/0952076718780537 Pollacci L, Sîrbu A, Giannotti F, Pedreschi D, Lucchese C, Muntean CI (2017) Sentiment Spreading: An Epidemic Model for Lexicon-Based Sentiment Analysis on Twitter. In: Esposito F, Basili R, Ferilli S (eds) Lisi (Éds.), AI*IA 2017 Advances in Artificial Intelligence, vol 10640. Springer International Publishing, pp 114–127. https://doi.org/10.1007/978-3-319-70169-1_9 Priem RL, Butler JE (2001a) Is the Resource-Based View a Useful Perspective for Strategic Management Research? Acad Manage Rev 26(1):22–40. https://doi.org/10.5465/amr.2001.4011928 Purdy M, Daugherty P (2016) Why artificial intelligence is the future of growth. Remarks at AI now: the social and economic implications of artificial intelligence technologies in the near term, 1–72 Sun TQ, Medaglia R (2019) Mapping the challenges of Artificial Intelligence in the public sector: Evidence from public healthcare. Government Inform Q 36(2):368–383. https://doi.org/10.1016/j.giq.2018.09.008 Susman GI, Evered RD (1978) Scientific Merits of Teece DJ, Pisano G, Shuen A (1997) Dynamic capabilities and strategic management. Strateg Manag J 18(7):509–533. https://doi.org/10.1002/(SICI)1097-0266(199708)18:7%3C509::AID-SMJ882%3E3.0.CO;2-Z Valérie D (2019) IA, la conquête de l’excellence technique. Revue De?F Nationale 820(5):123–130. Wang CL, Ahmed PK (2007) Dynamic capabilities: A review and research agenda. Int J Manage reviews 9(1):31–51 Wernerfelt B (1984) A resource-based view of the firm. Strateg Manag J 5(2):171–180. https://doi.org/10.1002/smj.4250050207 Wirtz BW, Weyerer JC, Geyer C (2019) Artificial Intelligence and the Public Sector Applications and Challenges. Int J Public Adm 42(7):596–615. https://doi.org/10.1080/01900692.2018.1498103 Yami S, Lehmann-Ortega L, Naro G (2008) Capacités dynamiques coopétitives: Le cas MSI dans la sous-traitance mécanique. XVIIth AIMS conference, Nice-Sophia Antipolis Zarrad A, Jaloud A, Alsmadi I The evaluation of the public opinion-a case study: Mers-cov infection virus in ksa. 2014 IEEE/ACM 7th International Conference on Utility and Cloud, Computing (2014) 664–670. https://ieeexplore.ieee.org/abstract/document/7027574/ Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8290777","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":561976956,"identity":"cdc2eabb-6ca3-4474-a3cd-9f6d02f17346","order_by":0,"name":"Montassar Bahri","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYBACAwYeBoYEBN8GiBkbDxClhQfCTwNpaSCshQGh5TCYxKvFnP3ssQ8Pc2zy7NmPP/7wc895u7Xth4G21NhE49Ji2ZOXPCNxW1oxD0+OmWTPs9vJ284kArUcS8ttwOWwAznGDInbDif2MOSwMfAcuJ1sdgCohbHhMG4t599AtfA/f/zxz4FzyWbnHxLQcgNmi0SCgTTPgQN2ZjcI2GI5A2xLWmLPjTdm0jIHkhPMbgBtScDjF3P+HGPGn9tsEtv70x9/fHPAzt7sfPrDBx9qbHBqwQCJYJUJxCoHAXtSFI+CUTAKRsHIAAD/qmbCVNpYpwAAAABJRU5ErkJggg==","orcid":"","institution":"Tunis El Manar University","correspondingAuthor":true,"prefix":"","firstName":"Montassar","middleName":"","lastName":"Bahri","suffix":""},{"id":561976961,"identity":"4170c5dd-2361-475d-8953-bb41891015f2","order_by":1,"name":"Mahmoud ZOUAOUI","email":"","orcid":"","institution":"University of Manouba","correspondingAuthor":false,"prefix":"","firstName":"Mahmoud","middleName":"","lastName":"ZOUAOUI","suffix":""}],"badges":[],"createdAt":"2025-12-05 21:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8290777/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8290777/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98779479,"identity":"eedd6332-17cb-42f7-a5eb-98269d9e3a11","added_by":"auto","created_at":"2025-12-22 12:30:23","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":384548,"visible":true,"origin":"","legend":"","description":"","filename":"ExperimentingwithanArtificialIntelligenceInitiativeinaPublicAdministrationContributionsfromanActionResearchApproach.docx","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/23ca0475b64c89627e80ddb9.docx"},{"id":98755034,"identity":"1d60f903-0bb1-4167-ab7f-34d994f61287","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4136,"visible":true,"origin":"","legend":"","description":"","filename":"e8ae83984f47498f8d444a4af9897212.json","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/e3f1a4a6391ac19155a786d3.json"},{"id":98778347,"identity":"216b69a7-a78c-4eed-987d-30e7835a6a73","added_by":"auto","created_at":"2025-12-22 12:29:11","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":129256,"visible":true,"origin":"","legend":"","description":"","filename":"e8ae83984f47498f8d444a4af98972121enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/8059ee20efd48cfb926975ff.xml"},{"id":98755041,"identity":"f4c10c76-d9b3-4123-a615-0aa360672245","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":242641,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/c5bca279f5d16819cd377fff.jpeg"},{"id":98779881,"identity":"c2ea2f4d-400a-4143-90f7-3585487deb2f","added_by":"auto","created_at":"2025-12-22 12:30:52","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":148316,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/67dc553ab2cf236da5b89a7d.jpeg"},{"id":98755038,"identity":"553a97c9-ecb2-4982-bb22-594034d4f18b","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":66181,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/76c6e57e830f5fbee1d77184.png"},{"id":98755043,"identity":"4ce2042a-ba81-4a17-9e0d-e6261ce70073","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110151,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/2a4af87a9944dcbeb64e92eb.png"},{"id":98778952,"identity":"54ab8e7b-4d59-47ab-845f-0fc065449d0c","added_by":"auto","created_at":"2025-12-22 12:29:51","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":46508,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/320179ad750066fd3ad59894.png"},{"id":98755046,"identity":"0faa2fa0-1f77-4416-baa6-225a38238a7c","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47002,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/845f1efd2923dc9baf0b209f.png"},{"id":98755049,"identity":"b4b0406c-5511-456a-9449-5e5c525e21cd","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32952,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/2f6755767b2cb8d24c563079.png"},{"id":98779836,"identity":"98e0bf8f-19d9-4828-8e5d-30704bfeedf1","added_by":"auto","created_at":"2025-12-22 12:30:48","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24022,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/fe23924b0b49f06ee4d21b12.png"},{"id":98755050,"identity":"590d918a-865c-4282-b9bd-8aacafa8df2d","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126864,"visible":true,"origin":"","legend":"","description":"","filename":"e8ae83984f47498f8d444a4af98972121structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/2bfff3bfe54c1b1cd7e3756f.xml"},{"id":98755048,"identity":"fc6c2d5c-fc70-4ce8-ae4d-1305170bd0d7","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"html","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":138224,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/3317748b9b061ba0e421d7b5.html"},{"id":98755035,"identity":"8e390612-933f-4cce-8c48-f163ab9fadbe","added_by":"auto","created_at":"2025-12-22 09:25:54","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":106765,"visible":true,"origin":"","legend":"\u003cp\u003eSentiment extraction and analysis steps (Alamoodi et al., 2021).\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/923e9748f3a7150583cf0939.jpg"},{"id":98777479,"identity":"63f2831b-a79a-416f-9554-5cb6690880c1","added_by":"auto","created_at":"2025-12-22 12:27:31","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":28856,"visible":true,"origin":"","legend":"\u003cp\u003eInvolvement of Different Actors in the Implementation of the “R” Tool\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/51507b5c24582f4b0c543687.jpg"},{"id":98778608,"identity":"c55a1dc8-9707-4af1-8b96-7c0b50e144ae","added_by":"auto","created_at":"2025-12-22 12:29:28","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":88215,"visible":true,"origin":"","legend":"\u003cp\u003eKey success factors for implementing an AI initiative in a public administration.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/f7353c6f897d7a805b9cf2bc.jpg"},{"id":100379808,"identity":"6b46e1f1-536b-4a05-9df3-6a43e4816a07","added_by":"auto","created_at":"2026-01-16 09:35:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1385777,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8290777/v1/92e67f97-1fe4-4b7c-a91b-fb287cdde4fb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Experimenting with an Artificial Intelligence Initiative in a Public Administration: Contributions from an Action Research Approach","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIn the context of transformations driven by the knowledge economy, managers are required to anticipate changes in their environment (Nutt \u0026amp; Backoff, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). AI has emerged as a major lever of this evolution, reshaping academic, economic, and geopolitical spheres (Parly, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Supported by the abundance of data, increased computational power, and the emergence of tools such as Matplotlib and NumPy (Anderson, 2014), AI now enables the forecasting of environmental developments and the automation of tasks requiring intelligence, speed, and precision.\u003c/p\u003e \u003cp\u003eInitially developed in the private sector, AI now attracts growing interest within public administrative organizations, where it opens new avenues for action. However, the literature emphasizes the need to consider potential risks and negative effects (Wirtz, 2019) and to favor simple, well-structured pilot projects. Few studies have simultaneously integrated the technical, human, and social dimensions involved in the implementation of an AI initiative.\u003c/p\u003e \u003cp\u003eMoreover, research remains limited regarding the value of AI in security institutions, even though it can potentially enhance their operational performance (Val\u0026eacute;rie D., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Scholars such as Mazzucchi and Guyonneau (2019) support this view, though without proposing a concrete implementation framework.\u003c/p\u003e \u003cp\u003eWhile AI applications in the security field are not new, their renewed relevance lies in the emergence of tools that enable more effective information exploitation. For the Tunisian Ministry of the Interior, the most relevant applications relate to data analysis and the detection of criminal threats.\u003c/p\u003e \u003cp\u003eThis study therefore seeks to design an AI initiative tailored to the Ministry\u0026rsquo;s analysis and research units, with the aim of improving the speed, accuracy, and reliability of their missions. Its objective is to identify the key success factors for such an initiative, propose appropriate methodologies, and support staff in their appropriation processes. Beyond the technological dimension, implementation requires an evolution of organizational values and behaviors (Santiso, 2019). An evaluation of the results will help assess both technical and behavioral changes, identify the challenges encountered, and formulate ethical recommendations that reconcile administrative values with AI-driven improvements.\u003c/p\u003e \u003cp\u003eBased on the foregoing, the research question guiding this study is as follows: How can a public administration implement an effective AI initiative within its information analysis and research services?\u003c/p\u003e \u003cp\u003eThis research adopts a constructivist stance within the action research paradigm. It mobilizes theory and practice to develop a conceptual framework and design an AI tool appropriate to the administrative context. The selection of the Tunisian Ministry of the Interiorgiven the strategic role of the security sectorenables assessment of the relevance of such an initiative and the proposal of a transferable integration strategy for other public administrations, through a qualitative approach grounded in case studies and data triangulation.\u003c/p\u003e \u003cp\u003eThe article presents a study structured into three phases, guided by Susman and Evered\u0026rsquo;s (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1978\u003c/span\u003e) model. The first phase involves diagnosing the perspectives of actors within the Tunisian Ministry of the Interior through ten interviews and participant observation, with the aim of identifying needs and challenges related to the implementation of an AI initiative. The second phase, based on an iterative learning process linking theory and practice, focuses on designing and launching the initiative in collaboration with the relevant actors, supported by ten additional interviews. The third phase evaluates the implementation in order to identify its benefits and the difficulties encountered at the material, organizational, and human levels, drawing on ten further interviews conducted with the same participants.\u003c/p\u003e"},{"header":"Conceptual Framework","content":"\u003cp\u003eDefining AI remains a complex task, given the breadth and interdisciplinary nature of its field of application. Limiting it to a single area of research would be reductive, as its foundations draw simultaneously from computer science, statistics, cognitive sciences, management, and organizational theory. From this perspective, it is relevant to mobilize several key concepts, particularly dynamic capabilities, knowledge management, and AI itself,in order to better understand the interactions that connect them. A joint analysis of these notions helps clarify the scope of the AI concept and identify the theoretical foundations that can inform its integration within organizations.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eAn Integrative Reading of Dynamic Capabilities in Light of Artificial Intelligence Technologies\u003c/h2\u003e \u003cp\u003eThe Resource-Based View (RBV), introduced by Wernerfelt (\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e1984\u003c/span\u003e), explains organizational performance through the nature and characteristics of the resources held by the firm. According to Barney (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e1991\u003c/span\u003e), resources can generate a competitive advantage when they are valuable and rare, and a sustainable advantage when they are also inimitable and non-transferable. Such resources enable organizations to seize opportunities, counter external threats, andthrough causal ambiguitymaintain an advantage that remains difficult for competitors to identify. However, several authors (Priem \u0026amp; Butler, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2001a\u003c/span\u003e, 2001b) have criticized the RBV for its static conception of performance, which is poorly suited to unstable environments. In response, the dynamic capabilities approach (Teece et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Wang \u0026amp; Ahmed, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) emerged as an evolutionary extension of the RBV, emphasizing the ability of organizations to adapt, integrate, and reconfigure their resources in the face of change. Subsequent empirical work (Yami et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) further developed this perspective by identifying the factors that support the development of dynamic capabilities and their contribution to organizational competitiveness.\u003c/p\u003e \u003cp\u003eWang and Ahmed (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) view resources as the foundation of any organization, while distinguishing them from capabilities. In this hierarchy, functional capabilities occupy the next level, corresponding to the organization\u0026rsquo;s ability to perform fundamental activities such as production, marketing, or distribution (Collis, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e1994\u003c/span\u003e). At the following level are core capabilities, which encompass the resources, skills, and knowledge that provide a sustainable competitive advantage (Leonard-Barton, 1992; Collis, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Wang \u0026amp; Ahmed, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Finally, dynamic capabilities stand at the top of this hierarchy, as they reflect an organization\u0026rsquo;s ability to transform and reconfigure its resources in response to change (Winter, 2003). These capabilities extend beyond mere adaptation and are rooted in structured organizational learning. Unlike ad hoc problem-solvingwhich relies on improvisationdynamic capabilities result from a continuous process of learning and preparation for change. As Josserand (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2007\u003c/span\u003e) argues, the dynamic recombination of resources is itself the outcome of an ongoing learning dynamic.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eA Theoretical Perspective on Artificial Intelligence as a Support for Knowledge Creation and Sharing\u003c/h3\u003e\n\u003cp\u003eAI contributes significantly to organizational knowledge management by facilitating the creation, storage, sharing, and application of knowledge. It enables the transformation of individual knowledge into collective and organizational knowledge, thereby enhancing organizations\u0026rsquo; capacity for learning and adaptation.\u003c/p\u003e\n\u003ch3\u003eKnowledge Creation\u003c/h3\u003e\n\u003cp\u003eAI, particularly deep learning, facilitates the discovery of new knowledge by detecting hidden patterns within data. It can predict trends (e.g., sales) and generate new insights from existing datasets. Practical applications, such as the analysis of millions of scientific publications, demonstrate that AI can identify previously unrecognized correlations, opening new avenues for research and innovation. In this way, AI helps organizations leverage big data and extract value from information that is often underutilized (with some studies suggesting that over 60% of data remains unused).\u003c/p\u003e\n\u003ch3\u003eKnowledge Storage and Retrieval\u003c/h3\u003e\n\u003cp\u003eAI enhances organizational memory by enabling the classification, organization, and retrieval of large volumes of data. Machine learning algorithms allow knowledge to be extracted and structured from multiple sources, as exemplified by recommendation systems (e.g., Gmail suggesting recipients). Consequently, AI makes knowledge bases more accessible, dynamic, and actionable.\u003c/p\u003e\n\u003ch3\u003eKnowledge Sharing\u003c/h3\u003e\n\u003cp\u003eIntelligent systems connect employees working on similar issues and foster collaborative intelligence. Platforms such as MITRE or collaborative tools like Slack and Yammer utilize AI to link individuals, stimulate collective creativity, and create a shared memory within teams.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eKnowledge Application\u003c/h2\u003e \u003cp\u003eAI facilitates the practical application of knowledge by tailoring it to specific needs and automating certain tasks. Intelligent assistants make relevant knowledge available in real time, thereby improving operational performance. For instance, Repsol has employed AI to drastically reduce unproductive time on its drilling sites through better data utilization. In customer service, platforms like Talla automate knowledge management to respond more rapidly to emerging needs.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eArtificial Intelligence\u003c/h3\u003e\n\u003cp\u003eBased on the various definitions of AI available in the literature, we have selected several that may help clarify how leaders perceive this concept. These definitions are summarized in the following table (Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e):\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDefinitions of the Concept of AI\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMarvin Lee Minsky (1956)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe development of computer programs that perform tasks currently accomplished more effectively by humans, as they require high-level mental processes such as perceptual learning, memory organization, and critical reasoning.\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eA technology capable of producing results similar to those generated by the human brain.\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHarry Shum (2018)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOperates only when there is a \u0026ldquo;vast amount of data; extraordinary computing power, particularly through the cloud; and groundbreaking algorithms based on deep learning.\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMcCarthy et al. (2006)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe learning process or any other feature of artificial intelligence can, in principle, be described so precisely that a machine can be made to simulate it.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRich et al. (2019)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIt is the way computers perform tasks that, at present, humans perform better.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRussel and Norvig, (2010)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI can be organized into four systems: systems that think like humans, systems that act like humans, systems that think rationally, and systems that act rationally.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRosa et al. (2016)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSoftware capable of learning, adapting, being creative, and solving problems.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThierer et al. (2017)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAn AI system is one that can undertake high-level operations; AI can function at or beyond human capabilities. This concept is divided into weak AI and strong AI.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eWithin the scope of this research, we adopt an operational definition of AI that is oriented toward the organizational and administrative context. :\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis definition primarily draws on the contributions of Russell and Norvig (2010), who distinguish between cognitive and rational approaches, as well as on McCarthy et al. (2006) regarding the simulation of human reasoning, while also integrating Harry Shum\u0026rsquo;s (2018) perspective on the importance of data and computational power as key enablers of contemporary artificial intelligence. It thus links the technological dimension of AI to an organizational purpose, consistent with the objectives of our action research focusing on the integration of an AI initiative within a public administration.\u003c/p\u003e \u003cp\u003eBased on this operational definition, AI should now be considered from the perspective of its integration into organizational contexts. Beyond its technical and algorithmic aspects, AI emerges as a strategic instrument capable of supporting transformations in management practices, decision-making, and value creation within organizations. This conception extends beyond simple task automation, engaging a deeper reconfiguration of organizational processes, human interactions, and data governance. From this standpoint, AI appears not only as an emerging technology but also as a driver of organizational change that fosters collective learning and administrative innovation.\u003c/p\u003e \u003cp\u003eWithin this framework, the introduction of AI into public administrations raises a range of managerial challenges. Public organizations, historically shaped by hierarchical structures and a culture of compliance, are now confronted with the need to adapt their structures and practices to fully leverage AI\u0026rsquo;s potential. This involves not only exploiting the operational benefits of the technology but also anticipating its impacts on public decision-making, competency management, transparency, and citizen trust. It is these multifaceted challenges that we address in the following section..\u003c/p\u003e\n\u003ch3\u003eChallenges in Implementing an AI Initiative within a Public Administration: Managerial Opportunities for Security Institutions\u003c/h3\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eApplications of AI in Public Administrative Organizations\u003c/h2\u003e \u003cp\u003eAI has generated considerable enthusiasm in the private sector in recent years, whereas the public sector still lags behind in AI adoption and deployment. Nevertheless, advances in AI offer a wide range of opportunities for its use within public organizations.\u003c/p\u003e \u003cp\u003eSeveral research studies have examined AI applications and challenges only in a fragmented and isolated manner. Scholars have synthesized scientific publications on AI to provide a comprehensive overview of its use in organizations. While public administrative organizations are increasingly experimenting with AI applications, their adoption remains less widespread than in the private sector. An analysis of multiple publications on this topic has identified ten major categories of AI applications within public organizations (Wirtz et al., 2018). These applications include AI-based knowledge management software, process automation systems, virtual agents, as well as predictive analytics and data visualization tools. They also encompass identity analysis systems, cognitive robotics and autonomous systems, recommendation engines, intelligent digital assistants, discourse analysis tools, and cognitive security and threat detection systems. Collectively, these technologies illustrate the diversity of possible AI uses in the administrative sphere and its potential to transform the operational modes of public institutions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eBenefits of AI Applications in Public Administrative Organizations\u003c/h2\u003e \u003cp\u003eAI not only drives technological innovation but also influences sociological, environmental, political, and administrative dimensions. Public organizations, as key actors in state development, are already witnessing AI integration across multiple areas of the public sector (Misuraca \u0026amp; Van Noordt, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Androutsopoulou et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Ojo et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Sun \u0026amp; Medaglia, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Although the pace and nature of this integration vary across countries, AI adoption is becoming progressively unavoidable in many government functions (De Sousa et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Moreover, the rapid evolution of AI technologies opens new opportunities for designing and implementing public policies (\u0026Ouml;nder \u0026amp; Saygili, 2018). Today, AI usage extends to various public services, as evidenced by Egger et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), who document multiple examples and adoption rates within the sector.\u003c/p\u003e \u003cp\u003eSeveral studies (Purdy \u0026amp; Daugherty, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Gartner \u0026amp; Hiebl, 2017) highlight the benefits of AI in public administrative organizations, including improved time management, development of self-learning and co-learning capabilities, and broader dissemination of innovation. They also emphasize the effectiveness of AI applications in reducing costs and time, enhancing accessibility, and promoting the inclusion of public services. Empirical evidence confirms a marked trend toward AI utilization in security, surveillance, and internal process optimization.\u003c/p\u003e \u003cp\u003eIn the security domain, Val\u0026eacute;rie D. (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) identifies several advantages of AI applications. These technologies enable more effective automatic detection of stealth targets and facilitate the exploitation of increasingly voluminous satellite data. They also provide valuable support to decision-makers by assisting them in making decisions within complex environments while maintaining operational speed. Furthermore, AI contributes to predictive maintenance, optimizing the availability of weapon systems, and plays a crucial role in detecting cyberattacks. It is also applied through robotic support on the battlefield or in aerial combat, as well as via human\u0026ndash;machine interaction functions in natural language. Additionally, AI enhances the performance of telecommunication networks through agile and automated routing and strengthens the realism of military training systems. Many of these applications also have relevant civilian uses, demonstrating the transversal and strategic significance of AI in the field of security.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePreparations for Implementing an AI Tool: Practical Framework (Data Collection No. 1)\u003c/h2\u003e \u003cp\u003eThe diagnostic phase is based on semi-structured interviews aimed at capturing the perspectives of leaders within the Tunisian Ministry of the Interior to identify obstacles, benefits, and challenges associated with implementing an AI initiative, while also collecting their proposals. The interview guide is designed to understand the perceived barriers, opportunities, and advantages of this initiative, thereby contributing to the refinement of the research conceptual framework. In total, ten ministry officials were interviewed in this context.\u003c/p\u003e \u003cp\u003eThe action research model adopted is that of Susman and Evered (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1978\u003c/span\u003e), recognized as one of the most widely used frameworks for describing this type of initiative (Baskerville, 1999). This model comprises five stages: diagnosis, action planning, action execution, evaluation, and knowledge generation, which are here applied to the implementation of an AI initiative within a public administration. The first two stagesdiagnosis and planningrequire continuous iteration between theory and fieldwork to verify and enrich the knowledge base through primary data collection with leaders of the Tunisian Ministry of the Interior. This data collection serves to refine the theoretical framework that will underpin the subsequent stages of AI implementation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eInitial Field Exploration: Refining Challenges and Preparing the Implementation Phase\u003c/h2\u003e \u003cp\u003eThis preliminary field exploration aims to determine the specific needs of the organizational context and guide the selection of the AI tool to be implemented within the public administration. It is positioned at this stage to refine understanding of the challenges and optimally prepare for the implementation phase.\u003c/p\u003e \u003cp\u003eAnalysis of the interviews reveals that leaders still rely heavily on traditional data analysis methods, while showing growing interest in innovative analytical tools. This openness reflects an explicit willingness to gradually introduce an AI initiative within their organizations, with the goal of improving decision-making practices and overall service performance.\u003c/p\u003e \u003cp\u003eInterviewed leaders clearly perceive the benefits associated with implementing an AI initiative. They particularly highlight improvements in processing quality and speed, as well as enhanced capabilities for data collection and analysis, leading to more relevant and timely outcomes. Beyond the technological dimension, these benefits are also linked to the development of internal competencies and improved knowledge management, positioning AI as a lever for organizational transformation and collective learning.\u003c/p\u003e \u003cp\u003eHowever, several barriers remain. The primary obstacles identified relate to the human dimension: resistance to change, perceived complexity, and lack of training. Staff may indeed struggle to adapt to new practices induced by AI. Therefore, adequate preparation is essential prior to any effective implementation. In this context, the role of leadership is central: leaders must ensure the technical readiness of the project, select and train staff capable of adopting these tools, and motivate them to support successful and sustainable AI adoption.\u003c/p\u003e \u003cp\u003eAccording to the interviewed leaders, three main approaches can be distinguished regarding the design and integration of AI systems. The first involves the development of generic AI algorithms by private, national, or international companies specializing in creating tools or platforms not intended for a specific application. The second corresponds to the implementation of ready-to-use AI systems, relying on existing software whose functionalities can be leveraged for certain specific tasks. Finally, the third approachadopted in this researchconsists of developing an AI system based on a concrete need and a specific problem, in this case a sentiment analysis tool. Accordingly, the phases of implementing an AI initiative can be grouped according to these three approaches, with their frequency of occurrence depending on organizational specificities, as illustrated in the following table (Table \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) :\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDifferent Phases of Design and Deployment Approaches According to the Officials\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGeneric AI Algorithms\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eReady-to-Use AI System\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCase-Specific AI System\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1. Investigation and Selection of the Company\u003c/p\u003e \u003cp\u003e2. Contact and Contract\u003c/p\u003e \u003cp\u003e3. AI Project Formulation: Requirements Gathering\u003c/p\u003e \u003cp\u003e4. AI System Development\u003c/p\u003e \u003cp\u003e5. Team Selection\u003c/p\u003e \u003cp\u003e6. Training\u003c/p\u003e \u003cp\u003e7. Tool Testing with Real Cases\u003c/p\u003e \u003cp\u003e8. Iteration and Evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1. AI Project Formulation: Requirements Gathering\u003c/p\u003e \u003cp\u003e2. Ensuring Confidentiality (IT Security)\u003c/p\u003e \u003cp\u003e3. Selection of Tools Adaptable to the Service\u003c/p\u003e \u003cp\u003e4. Team Selection\u003c/p\u003e \u003cp\u003e5. Training\u003c/p\u003e \u003cp\u003e6. Tool Testing with Real Cases\u003c/p\u003e \u003cp\u003e7. Iteration and Evaluation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1. AI Project Formulation: Requirements Gathering\u003c/p\u003e \u003cp\u003e2. Selection of Tools Adaptable to the Use Case\u003c/p\u003e \u003cp\u003e3. Team Selectio\u003c/p\u003e \u003cp\u003e4. Specialized Training\u003c/p\u003e \u003cp\u003e5. Tool Testing with Real Cases\u003c/p\u003e \u003cp\u003e6. Iteration and Evaluation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAt the conclusion of this phase, the interviewees opted for the implementation of a sentiment analysis tool, considering that it would enable a better understanding of employees\u0026rsquo; perceptions and opinions, identify friction points within internal processes, and support data-driven decision-making based on reliable qualitative insights.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eSentiment Analysis\u003c/h2\u003e \u003cp\u003eThe literature employs various terms to refer to sentiment analysis (SA), such as opinion mining, emotion analysis, subjectivity analysis, or evaluation of phenomena. According to Medhat et al. (2014), the most commonly used terms are opinion mining and sentiment analysis, which are considered equivalent and fall under the subfield of subjectivity analysis. These authors define sentiment analysis as the computational processing of opinions, emotions, and judgments expressed in a text, situated within the broader field of text mining research. However, some studies conflate the notions of sentiment and opinion: according to the Merriam-Webster dictionary, an opinion is a judgment or evaluation on a specific subject, whereas sentiment refers to an attitude or judgment influenced by emotion.\u003c/p\u003e \u003cp\u003eFurthermore, the rise of social mediacharacterized by ease of communication and idea-sharing within communities (Kim, Sin \u0026amp; YooLee, 2014)has made these platforms among the most dynamic and influential information systems today (Lin, Yang \u0026amp; Li, 2020). The psychological or cognitive effects of social media data on user behavior cannot be fully measured (Crawford, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Opinions and ideas shared on these platforms are varied and may sometimes be biased or manipulative (Jansen, Sobel \u0026amp; Cook, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). While widely shared posts can serve as valuable sources of information, their interpretation remains complex due to the diversity of emotions expressed, including negation, sarcasm, sadness, joy, anger, surprise, or concern (Ji et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Ali et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Zarrad et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Nevertheless, exploring these emotions constitutes a strategic tool for understanding public dynamics and guiding political or economic decision-making (Chung, He \u0026amp; Zeng, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOpinion mining and sentiment analysis in the context of big data have become increasingly useful, as they contribute to understanding human emotions by enabling the tracking of behavioral patterns when users engage with social media applications (Ji et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eSentiment Analysis Process\u003c/h2\u003e \u003cp\u003eAccording to Alamoodi et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), the sentiment analysis process consists of four main stages (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). It begins with source selection, which involves determining the platform from which sentiments will be extracted, such as Facebook, Twitter, YouTube, Instagram, or Threads (Chung et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). The next stage is data collection, during which information is gathered using hashtags (#) (Gayo-Avello et al., 2013) or keywords (Deng, Tang \u0026amp; Huang, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), in various forms such as posts, comments, news articles, or texts (Chung et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe third stage is preprocessing, where collected data are cleaned and normalized to remove irrelevant elements such as repeated letters, stop words, or typographical errors, while extracting the necessary linguistic features (Baker et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Pollacci et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Finally, the sentiment analysis stage utilizes the preprocessed data, either automatically or manually, to identify the emotions expressed and their frequency (Deng et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). This structured process ensures a more precise and reliable understanding of opinions conveyed through digital content.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAccording to Andrea Esuli and Fabrizio Sebastiani (2006), sentiment analysis also referred to as opinion mining has a wide range of applications. It enables tracking public attitudes toward a political party based on data from social media, electronic newspapers, or blogs. Similarly, it can be used to assess consumer perceptions of a product by automatically classifying online reviews. Rooted in natural language processing and text analysis, this technique aims to explore and categorize opinions in an automated manner. Its primary applications span multiple domains. In decision-making, opinions gathered from social media provide valuable behavioral insights that inform strategic organizational choices. In public policy development, online citizen opinion analysis can guide policymakers in designing policies better aligned with societal expectations. It is also particularly useful for detecting hate speech, facilitating the automated monitoring of offensive, provocative, or harmful content on blogs, news sites, and social platforms. Finally, in marketing research, this approach contributes to the study of products, services, and consumption trends, while also enabling the analysis of public attitudes toward government policies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eAction Research as an Operational Framework\u003c/h2\u003e \u003cp\u003eWe adopted the action research model of Susman and Evered (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1978\u003c/span\u003e), widely applied in management sciences (Baskerville, 1999). This model comprises five stages: diagnosis, planning, action, evaluation, and learning. It provides a relevant framework for studying AI implementation as it combines knowledge generation with organizational transformation. The method promotes collaboration between researchers and practitioners, while allowing continuous adaptation of actions to the specificities of the administrative context.\u003c/p\u003e \u003cp\u003eThe research employed multiple intervention cycles, combining semi-structured interviews and participant observation with officials from the Tunisian Ministry of the Interior. These iterations allowed refinement of the conceptual framework and the real-world testing of a sentiment analysis AI tool as a decision-support mechanism and a means of modernizing public services.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eAlignment of Paradigm, Method, and Purpose\u003c/h2\u003e \u003cp\u003eThe articulation between the constructivist paradigm and the action research approach confers both epistemological and methodological coherence to this study. Constructivism enables a situated and shared understanding of change, considering organizational reality as the product of interactions and interpretations of actors. The abductive approach, in turn, facilitates continuous interplay between theory and practice, allowing knowledge to be built from experience while being enriched by relevant conceptual frameworks. Finally, action research fosters a reflective dynamic in which knowledge informs action, and action, in turn, generates new knowledge. This alignment between paradigm, method, and purpose enhances the scientific and operational relevance of the adopted approach.\u003c/p\u003e \u003cp\u003eConsequently, this study does not merely aim to describe a phenomenon but seeks to produce useful, transferable knowledge applicable to other public organizations. By integrating the researcher\u0026rsquo;s reflective stance and the involvement of field actors, the proposed approach contributes to strengthening organizational learning capacity and understanding the conditions for successful AI integration in the public sector.\u003c/p\u003e \u003c/div\u003e"},{"header":"Methodological Choices","content":" \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003cp\u003eA qualitative approach was deemed appropriate to analyze AI implementation within a security-focused public administration. This choice is justified by the nature of the phenomenon, which involves human, organizational, and managerial dimensions that are difficult to quantify. Consistent with Collerette (1996) and Hlady-Rispal (2002), the qualitative approach aims to understand the complexity of an evolving social system and to capture actors\u0026rsquo; perceptions, barriers, and levers.\u003c/p\u003e \u003cp\u003eData collection relied on semi-structured interviews complemented by direct observation of organizational dynamics. Three interview guides were developed based on the literature review to ensure coherence between research objectives and the information collected. The first guide focused on the initial diagnosis of AI implementation, exploring perceived barriers, benefits, and challenges. The second guide addressed the implementation of a sentiment analysis tool, capturing technical and organizational integration aspects. Finally, the third guide aimed to evaluate the outcomes and capitalize on knowledge gained, producing lessons transferable to future similar initiatives. These guides were pre-validated by participants to ensure trust and improve data quality.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eField Context\u003c/h2\u003e \u003cp\u003eThe study was conducted within the Directorate General of Special Services of the Tunisian Ministry of the Interior, where three sub-directorates were selected due to their strong commitment to digital transformation. Sub-directorate A, focused on human development and organizational agility, leverages digital transformation to enhance analysts\u0026rsquo; practices and efficiency. Sub-directorate B, despite possessing strong technical competencies, faces certain administrative constraints and a lack of analytical tools, limiting its innovation potential. Sub-directorate C, primarily oriented toward social media data collection and analysis, seeks to improve the reliability and performance of its analyses through AI integration. This organizational diversity allowed observation of different AI adoption dynamics according to the structural and cultural contexts of each unit, providing representative insight into the challenges faced in the Tunisian administration.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eData Collection and Processing\u003c/h2\u003e \u003cp\u003eData were collected between 2019 and 2024 within an action research framework. The researcher, embedded within the studied organization, conducted ten exploratory interviews followed by twenty semi-structured interviews during the implementation and evaluation phases. Interviews, averaging two hours each, involved leaders, administrative and technical managers, and the central director. In total, over 42 hours of formal interviews and several observation sessions were conducted.\u003c/p\u003e \u003cp\u003eInterview settings were arranged to promote spontaneity and trust. The researcher\u0026rsquo;s role, oscillating between consultant and observer, minimized bias while fostering contextualized understanding. Probing questions (\u0026ldquo;Could you clarify?\u0026rdquo;, \u0026ldquo;Could you confirm?\u0026rdquo;) facilitated validation and coherence of the collected narratives.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eData Analysis\u003c/h2\u003e \u003cp\u003eData were transcribed, coded, and analyzed using MaxQDA, following an iterative coding process. An initial coding grid was developed in Excel and refined after analyzing sub-directorates A and B to better capture key dimensions of organizational change, technological barriers, and human levers. The final grid, applied to sub-directorate C and the central director, yielded structured thematic categories: success conditions for AI initiatives, human and organizational factors, leadership roles, and impacts on managerial practices.\u003c/p\u003e \u003cp\u003eTriangulation of sources (interviews, observations, internal documents) strengthened the validity of the results and enabled an integrated reading of transformation dynamics. The analysis underscores the importance of human capital, training, and change management support for the successful implementation of AI initiatives in the public sector.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eWe have chosen to present the results of the first phase of our research, corresponding to the diagnostic and action planning stages, alongside the exploration and enrichment of the theoretical framework. This preliminary phase allowed for the identification and supplementation of certain missing concepts. In particular, the analysis of exploratory interviews highlighted the necessity of integrating the concept of sentiment analysis, which was mentioned by the managers during the very first interview. This methodological choice aims to enhance the coherence between the conceptual framework and the empirical research process. The following section then presents the results of the second phase, concerning the implementation of the tool, as well as those of the third phase, focusing on the evaluation of the action.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResults of the Second Phase: Implementation of a Sentiment Analysis Tool within the Tunisian Ministry of the Interior\u003c/b\u003e \u003c/p\u003e \u003cp\u003eBefore initiating the implementation of an AI tool across three sub-directorates, we sought to anticipate the expectations and potential changes within these units. The primary objective was to understand the perceived utility of the sentiment analysis tool and to identify doubts regarding its actual contributions, thereby aligning its potential with the operational and strategic needs of the examined entities.\u003c/p\u003e \u003cp\u003eThe deployment of the sentiment analysis tool primarily aimed to optimize analysts\u0026rsquo; and operational actors\u0026rsquo; working time, while improving the overall relevance of organizational activities. Initially, the project had no formal status or dedicated teams, as it originated from research initiatives and preliminary field interviews. The choice of the R programming language as a technical support tool was driven by several decisive advantages. R allows for results comparable to those obtained with paid software while integrating both backend (programming) and frontend (interface) functionalities within a single environment. It also enables all types of analyses to be performed freely using customizable scripts, without any licensing constraints. Therefore, adopting R met a dual requirement of efficiency and resource optimization, while fostering gradual appropriation of AI within a transforming administrative context.\u003c/p\u003e \u003cp\u003eTo address the lack of technical skills among interviewees, a detailed guide was developed explaining the research objectives, implementation steps, and technical aspects of using the tool. This guide was distributed to the central and general directorates of the Tunisian Ministry of the Interior with the approval of the Director General of National Security.\u003c/p\u003e \u003cp\u003eThe primary objective of the R tool was to simplify sentiment analysis, thereby optimizing analysts\u0026rsquo; work and addressing the challenges posed by the absence of suitable tools for processing and interpreting textual data in a security context. As deployment progressed, the range of actors involved in the project diversified, and their roles gradually became defined.\u003c/p\u003e \u003cp\u003eThe main actors involved in the process and their respective contributions are illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe figure above illustrates that the directors of the sub-directorates played a central role in identifying and formalizing organizational needs, while coordinating teams and maintaining liaison with hierarchical authorities. Specific managers served as points of contact between the directors and the analytical units, ensuring the transmission of directives and supervising the various stages of data collection. Analysts, as the primary users of the tool, actively participated in both the collection and analysis of sentiments, as well as in the practical exploitation of the results. Operational staff collaborated with analysts to guide data selection and support the tool\u0026rsquo;s testing phases. Finally, the researcher provided technical and methodological support by organizing specialized training on the R programming language, installing the tool on analysts\u0026rsquo; workstations, and guiding them step by step in its use. This distribution of roles ensured a collaborative and gradual implementation of the sentiment analysis approach.\u003c/p\u003e \u003cp\u003eAll actors played dual roles, both in the design and deployment of the tool and as end-users, contributing simultaneously to its improvement and to the enhancement of the teams\u0026rsquo; analytical capabilities.\u003c/p\u003e \u003cp\u003eThe introduction of the AI-based tool raised several questions regarding its actual utility and added value for the activities of the concerned units. AI tools, by enabling the processing of large volumes of data and synthesizing information, offer significant potential to facilitate and inform decision-making. Early feedback, however, highlighted several tangible benefits. The tool improved analysts\u0026rsquo; performance and work quality, while contributing to the enhancement of human capacities by helping users better execute their daily tasks. Moreover, it promoted a more efficient exploitation of textual data, transforming it into relevant and actionable knowledge, particularly valuable in the security domain. The integration of this AI solution not only optimized analytical processes but also strengthened collective capacity to generate more nuanced and responsive decision intelligence.\u003c/p\u003e \u003cp\u003eThus, the implementation phase highlighted not only the anticipated benefits of the tool but also the conditions necessary for its successful adoption and integration within the examined sub-directorates.\u003c/p\u003e \u003cp\u003eFollowing the implementation of the R-based sentiment analysis tool, several key success factors were identified to ensure the effective deployment of an AI initiative within a public administration. The process was structured around several interdependent steps. The first step consisted of collecting needs and data to clearly define organizational expectations and align them with the practical usefulness of AI tools. The second step involved selecting the appropriate AI tool based on the identified needs and on criteria such as effectiveness, functionality, cost, and ease of use. The third step focused on establishing a testing team composed of motivated and competent members responsible for piloting the implementation and monitoring the system. Comprehensive training was then provided to enable the team to fully exploit the tool\u0026rsquo;s potential through hands-on exercises. Finally, a continuous iteration and evaluation phase ensured the progressive adaptation of the tool to the organization\u0026rsquo;s specific needs and to the diversity of its application areas, including social, economic, political, educational, and counter-terrorism domains. This gradual and reflective approach underpins the sustainability and relevance of the AI initiative undertaken.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eTechnical Challenges and Learning Opportunities\u003c/h2\u003e \u003cp\u003eThe technical challenges encountered, such as managing dependencies among R packages, although initially perceived as obstacles, actually served as learning levers. They fostered the development of proactive strategies, continuous improvement of analytical practices, and the transfer of skills within the team. These technical difficulties thus became opportunities to strengthen both the efficiency and mastery of the tools, directly contributing to the success of the AI strategy.\u003c/p\u003e \u003cp\u003eBased on our observations, implementing an AI initiative within an administrative organization entails several challenges related to training data. Data access varies across domains (security, economy, etc.), which can create difficulties in obtaining a significant data volume, sometimes leading to non-transparent acquisition methods. Data design must account for quantity, completeness, and feature quality, while data utilization requires prior cleaning to ensure analytical reliability. Consequently, the AI system cannot learn or improve autonomously without human intervention.\u003c/p\u003e \u003cp\u003eAnother critical challenge is the network effect: the more relevant data the AI integrates, the more effective it becomes. Data collection and cleaning phases require significant effort, and the quality of analytical results directly depends on the number of users and the amount of data exploited. We also observed that qualitative data analysis requires a team of at least four people to ensure effective implementation. Observed limitations include a perception of \u0026ldquo;handcrafted\u0026rdquo; design, concerns about reliability and reproducibility, and a sometimes overly technology-centered approach rather than one guided by actual organizational needs. These findings underscore the importance of aligning design with specific contextual requirements, particularly in security-related domains.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResults of the Third Phase: Evaluation and Interpretation \u0026ndash; AI at the Heart of the Administrative Organization\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo evaluate the implementation of an AI initiative within three administrative sub-directorates, we employed thematic content analysis. This approach allowed us to explore qualitative data from interviews and observations to understand the effects of the R-based sentiment analysis tool on organizational practices and anticipated changes.\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eEnhancement of Competence Management\u003c/h2\u003e \u003cp\u003eThe deployment of the R tool enabled analysts to acquire new theoretical and practical knowledge, significantly improving previously heterogeneous analytical practices. The action-research process involved formalizing explicit procedures, ensuring alignment between actors\u0026rsquo; expectations and the AI tool\u0026rsquo;s potential. Field evaluation revealed that directors began reconfiguring resources and competencies within the sub-directorates, activating dynamic capabilities to implement rapid changes.\u003c/p\u003e \u003cp\u003eKey steps in competence management can be summarized in four main axes: identification of skills and talents to pinpoint human resources capable of contributing effectively to organizational objectives; discovery and support of talents to foster their development and valorization; exploitation of skills by mobilizing them optimally to meet strategic and operational needs; and feedback integration to continuously adjust and improve competence management, promoting ongoing learning and organizational development.\u003c/p\u003e \u003cp\u003eThese observations confirm the dynamic capabilities model (Labrouche 2014), integrating organizational learning, strategic adaptation, and the creation of distinctive resources to respond to technological and organizational changes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003eNew Vision of Knowledge Management\u003c/h2\u003e \u003cp\u003eContent analysis showed that introducing the R tool fostered a true culture of knowledge production and management among analysts. Its use encouraged quality-oriented behavior and continuous learning, generating reliable and actionable insights for decision-making. The knowledge management cycle observed can be divided into several complementary steps: cataloging, i.e., exhaustive collection of existing knowledge, including available resources and analysts\u0026rsquo; know-how; structuring, to organize data for easier exploitation; verification, to ensure accuracy through collaboration with directors and the researcher; dissemination, to make knowledge accessible to relevant actors; and continuous optimization, to enhance practices and results based on feedback and new data, thus reinforcing organizational learning dynamics.\u003c/p\u003e \u003cp\u003eThis process transformed analysts\u0026rsquo; behavior, fostering continuous learning and the production of high-quality analytical knowledge.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003eOrganizational Utility of the Tool\u003c/h2\u003e \u003cp\u003eThe R tool demonstrated significant organizational value within sub-directorates A, B, and C. Directors were able to reassess and adapt strategies, integrating an organizational culture based on innovation and responsiveness to administrative evolution. The tool also enabled the acquisition of new management capabilities, reducing errors and improving analytical output quality.\u003c/p\u003e \u003cp\u003eImpact analysis revealed multiple benefits within the organization. The tool enhanced effectiveness in achieving objectives while ensuring efficiency through optimal resource utilization. It facilitated intelligent exploration, enabling discovery and investigation of new information, and optimal use of collected data. Furthermore, it contributed to the continuous development of analysts\u0026rsquo; skills while ensuring the relevance and reliability of produced results. Additionally, it supported timely decision-making and the creation of clear, innovation-based strategies. In the security domain, the tool facilitated preventive policing, risk anticipation, and a deeper understanding of root causes, helping to prevent recurrence. These combined effects illustrate the strategic and operational value that AI can provide within a complex organizational context.\u003c/p\u003e \u003cp\u003eThus, the R tool not only enhanced individual analytical skills but also reinforced organizational culture, generating short-, medium-, and long-term benefits for the organization.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Discussion of Results","content":" \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003cp\u003eThis article explored the transformations induced by the design, deployment, and challenges associated with implementing an AI initiative within an administrative organization, particularly in the security domain. The three successive empirical studies conducted across three sub-directorates highlighted issues related to AI integration, the use and impact of AI tools on the quality of analytical outcomes, and the practical design and deployment process of the R-Studio tool, tailored to the specific needs of the sub-directorates under study.\u003c/p\u003e \u003cp\u003eThe main objective in designing the R tool was to train and guide analysts to adapt to its capabilities, develop their skills, and produce accurate and relevant results. Participants were involved either in identifying and formalizing requirements or directly contributing to the tool\u0026rsquo;s technical development. The design phase also included deployment in operational contexts, continuous optimization, and process improvement based on feedback from directors and managers.\u003c/p\u003e \u003cp\u003eSeveral challenges were encountered before, during, and after implementation. Prior to deployment, identifying organizational needs proved complex, as actors were not always familiar with AI, and the administrative organization had no prior experience. Administrative instability complicated the selection of project leaders, and bureaucratic behaviors slowed necessary authorizations. Additionally, no strategy existed to evaluate internal competencies or to raise analyst awareness of AI tools. During implementation, hardware constraints, such as IT system incompatibilities and performance limitations, hindered tool usage. Analysts faced difficulties executing code and managing regular updates, while data security and confidentiality required strict protocols. Post-implementation, the absence of a strategy for technical adjustments and continuous training limited tool optimization, and analysts struggled to correctly interpret some outputs.\u003c/p\u003e \u003cp\u003eThe impacts of AI introduction were multifaceted. The tool improved performance and work quality, supporting decision-making, particularly in the analysis of social phenomena. However, these benefits are not universal and depend heavily on context, activities, and usage conditions. AI can sometimes generate inefficiencies or increased workloads if its limitations are not understood, and there is a risk of overconfidence or unrealistic expectations regarding the tool.\u003c/p\u003e \u003cp\u003eThe study showed that AI deployment mobilizes two major organizational levers. Dynamic capabilities manifested through the identification of opportunities provided by AI, progressive appropriation via experimentation, integration of tools into administrative tasks, and adjustment of workflows to fully exploit AI potential. Knowledge management was essential for structuring, externalizing, and internalizing tacit knowledge, consolidating best practices and operational procedures within work routines. The use of action research as an integrative framework allowed observation of these dynamics in real time, adjustment of interventions, and documentation of how dynamic capabilities and knowledge management were mobilized within a specific security context.\u003c/p\u003e \u003cp\u003eFinally, ensuring the sustainability of the R tool remains a challenge. Its utility varies across sub-directorates and usage contexts, and its real relevance for decision-making may be questioned due to the confidentiality of information processed. Administrative instability and leadership transitions are also significant challenges. These elements highlight that real organizational performance benefits are difficult to measure, emphasizing the need for ongoing monitoring and a clear strategy to ensure sustained adoption of the AI tool.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis article examined the design and implementation of an AI initiative within an administrative organization, aiming to address the central question: how can a research and analysis service deploy an effective AI initiative? The study adopted a constructivist approach and action research as an integrative framework, combining field diagnostics, the design and deployment of an AI-based sentiment analysis tool, and evaluation of its implementation.\u003c/p\u003e \u003cp\u003eKey results identified success factors such as data collection focused on organizational needs, appropriate selection of the tool and pilot team, adequate training, experimentation with real cases, and regular iterations for evaluation and adjustment. These factors highlighted not only observable impacts on analytical quality and work methods but also subtler managerial effects related to theoretical concepts such as dynamic capabilities and knowledge management.\u003c/p\u003e \u003cp\u003eMethodologically, the abductive approach allowed addressing initial questions while deeply exploring the implications of tool deployment. Integration of Susman \u0026amp; Evered\u0026rsquo;s (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1978\u003c/span\u003e) action research model facilitated knowledge acquisition and capitalization by actors, creating an organizational learning process. The multiple-case study design, covering the phases of diagnosis, planning, action, evaluation, and knowledge specification, enabled an in-depth analysis of AI-induced transformations from design to final assessment.\u003c/p\u003e \u003cp\u003eEmpirical contributions indicate that AI integration in administrative organizations can generate significant structural and organizational changes, improve analytical practices, and raise awareness among actors about the continuously evolving field. Tool evaluation involved the central director, directors, and managers of the three sub-directorates, allowing both theoretical and practical insights. This approach highlighted essential elements for successful implementation: defining processes, anticipating challenges, and considering impacts on activities, roles, and organization.\u003c/p\u003e \u003cp\u003eThe study also provides original contributions by exploring AI adoption in the Tunisian public sector, a relatively under-researched area, particularly in developing country administrations. It shows how dynamic capabilities and knowledge management enable organizational actors to progressively appropriate AI technologies by adapting resources and processes to technological changes. Initial AI deployments within the Tunisian Ministry of the Interior revealed significant gains in data processing and decision-making, emphasizing the importance of adequate preparation, continuous training, leadership support, and administrative flexibility.\u003c/p\u003e \u003cp\u003eFinally, the research acknowledges certain limitations, including the mainly declarative nature of the first phase based on semi-structured interviews and the qualitative scope limited to three organizational units, which restricts generalizability. Future research could expand the sample, integrate a quantitative approach, and replicate the study across different contexts and AI tools to identify similarities and divergences, improving the proposed analytical framework. This approach could guide the design and deployment of AI systems in other administrative organizations and enhance understanding of AI\u0026rsquo;s organizational and managerial impacts.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval statement This research did not require formal approval from an institutional ethics committee, as it did not involve medical experimentation or sensitive personal data. The study was conducted in accordance with internationally recognized ethical principles for social science research. Participation was voluntary, participants were informed of the objectives of the study, and informed consent was obtained. All data were anonymized, and confidentiality was strictly maintained.\u003c/p\u003e\u003cp\u003e \u003ch2\u003eClinical trial number\u003c/h2\u003e \u003cp\u003enot applicable.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThe authors received no financial support for the research, authorship, and/or publication of this article.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthorship Contribution StatementM.B. conducted this research as part of his doctoral thesis and wrote the main manuscript. M.Z. supervised the work, provided methodological guidance, and contributed to the revision and refinement of the manuscript. All authors reviewed and approved the final version of the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlamoodi AH, Zaidan BB, Zaidan AA, Albahri OS, Mohammed KI, Malik RQ, Almahdi EM, Chyad MA, Tareq Z, Albahri AS (2021) Sentiment analysis and its applications in fighting COVID-19 and infectious diseases: A systematic review. Expert Syst Appl 167:114155. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciencedirect.com/science/article/pii/S0957417420308988\u003c/span\u003e\u003cspan address=\"https://www.sciencedirect.com/science/article/pii/S0957417420308988\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli K, Dong H, Bouguettaya A, Erradi A, Hadjidj R (2017) Sentiment analysis as a service: A social media based sentiment analysis framework. 2017 IEEE international conference on web services (ICWS), 660\u0026ndash;667. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ieeexplore.ieee.org/abstract/document/8029820/\u003c/span\u003e\u003cspan address=\"https://ieeexplore.ieee.org/abstract/document/8029820/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnderson N, Potočnik K, Zhou J (2014) Innovation and Creativity in Organizations: A State-of-the-Science Review, Prospective Commentary, and Guiding Framework. J Manag 40(5):1297\u0026ndash;1333. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0149206314527128\u003c/span\u003e\u003cspan address=\"10.1177/0149206314527128\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndroutsopoulou A, Karacapilidis N, Loukis E, Charalabidis Y (2019) Transforming the communication between citizens and government through AI-guided chatbots. Government Inform Q 36(2):358\u0026ndash;367. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciencedirect.com/science/article/pii/S0740624X17304008\u003c/span\u003e\u003cspan address=\"https://www.sciencedirect.com/science/article/pii/S0740624X17304008\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAraujo T, Helberger N, Kruikemeier S, De Vreese CH (2020) AI we trust? Perceptions about automated decision-making by artificial intelligence. AI Soc 35(3):611\u0026ndash;623. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00146-019-00931-w\u003c/span\u003e\u003cspan address=\"10.1007/s00146-019-00931-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaker QB, Shatnawi F, Rawashdeh S, Al-Smadi M, Jararweh Y (2020) Detecting epidemic diseases using sentiment analysis of arabic tweets. J Univers Comput Sci 26(1):50\u0026ndash;70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.academia.edu/download/72331153/pdf.pdf\u003c/span\u003e\u003cspan address=\"https://www.academia.edu/download/72331153/pdf.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarney J (1991) Firm Resources and Sustained Competitive Advantage. J Manag 17(1):99\u0026ndash;120. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/014920639101700108\u003c/span\u003e\u003cspan address=\"10.1177/014920639101700108\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaskerville RL, Wood-Harper AT (1996) A critical perspective on action research as a method for information systems research. J Inform Technol 11(3):235\u0026ndash;246. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/026839696345289\u003c/span\u003e\u003cspan address=\"10.1080/026839696345289\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaskerville R, Smithson S (1995) Information technology and new organizational forms: Choosing chaos over panaceas. Eur J Inform Syst 4(2):66\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1057/ejis.1995.8\u003c/span\u003e\u003cspan address=\"10.1057/ejis.1995.8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBernard F (2004) Constructivisme et sciences de l\u0026rsquo;organisation. De l\u0026rsquo;alternative au pluralisme \u0026eacute;pist\u0026eacute;mologique \u0026laquo;limit\u0026eacute;\u0026raquo;. Communication Langages 139(1):27\u0026ndash;40. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.persee.fr/doc/colan_0336-1500_2004_num_139_1_3249\u003c/span\u003e\u003cspan address=\"https://www.persee.fr/doc/colan_0336-1500_2004_num_139_1_3249\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCharreire S, Huault I (2001) Coh\u0026eacute;rence \u0026eacute;pist\u0026eacute;mologique et recherche en management strat\u0026eacute;gique\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung W, He S, Zeng DD (2015) eMood: Modeling Emotion for Social Media Analytics on Ebola Disease Outbreak Research-in-Progress\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollis DJ (1994) Research Note: How Valuable are Organizational Capabilities? Strateg Manag J 15(S1):143\u0026ndash;152. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/smj.4250150910\u003c/span\u003e\u003cspan address=\"10.1002/smj.4250150910\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrawford K (2009) Following you: Disciplines of listening in social media. Continuum 23(4):525\u0026ndash;535. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10304310903003270\u003c/span\u003e\u003cspan address=\"10.1080/10304310903003270\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Sousa WG, de Melo ERP, Bermejo PHDS, Farias RAS, Gomes AO (2019) How and where is artificial intelligence in the public sector going? A literature review and research agenda. Government Inform Q 36(4):101392\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng L, Yu D (2014) Deep learning: Methods and applications. Foundations and trends\u0026reg; in signal processing, 7(3\u0026ndash;4) 197\u0026ndash;387. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nowpublishers.com/article/Details/SIG-039\u003c/span\u003e\u003cspan address=\"https://www.nowpublishers.com/article/Details/SIG-039\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng Xlong, Tang YQ, Huang YH (2015) Opinion mining for emergency case risk analysis in spark based distributed system. Proceedings of the 1st ACM SIGSPATIAL International Workshop on the Use of GIS in Emergency Management, 1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2835596.2835608\u003c/span\u003e\u003cspan address=\"10.1145/2835596.2835608\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDwivedi YK, Rana NP, Jeyaraj A, Clement M, Williams MD (2019) Re-examining the Unified Theory of Acceptance and Use of Technology (UTAUT): Towards a Revised Theoretical Model. Inform Syst Front 21(3):719\u0026ndash;734. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10796-017-9774-y\u003c/span\u003e\u003cspan address=\"10.1007/s10796-017-9774-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEgger F, Benzing V, Conzelmann A, Schmidt M (2019) Boost your brain, while having a break ! The effects of long-term cognitively engaging physical activity breaks on children\u0026rsquo;s executive functions and academic achievement. PLoS ONE 14(3):e0212482. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://journals.plos.org/plosone/article?id=10.1371/journal.pone.0212482\u003c/span\u003e\u003cspan address=\"https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0212482\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsuli A, Sebastiani F, Abbasi A (2010) Sentiment Quantification. IEEE Intell Syst 25(4):72\u0026ndash;75\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJansen BJ, Sobel K, Cook G (2010) Gen X and Ys attitudes on using social media platforms for opinion sharing. CHI \u0026rsquo;10 Extended Abstracts on Human Factors in Computing Systems. 3853\u0026ndash;3858. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/1753846.1754068\u003c/span\u003e\u003cspan address=\"10.1145/1753846.1754068\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi R, Cao D, Zhou Y, Chen F (2016) Survey of visual sentiment prediction for social media analysis. Front Comput Sci 10(4):602\u0026ndash;611. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11704-016-5453-2\u003c/span\u003e\u003cspan address=\"10.1007/s11704-016-5453-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi X, Chun SA, Geller J (2016) Knowledge-Based Tweet Classification for Disease Sentiment Monitoring. In W. Pedrycz \u0026amp; S.-M. Chen (\u0026Eacute;ds.), Sentiment Analysis and Ontology Engineering: An Environment of Computational Intelligence (pp. 425\u0026ndash;454). Springer International Publishing. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-319-30319-2_17\u003c/span\u003e\u003cspan address=\"10.1007/978-3-319-30319-2_17\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJi X, Chun SA, Wei Z, Geller J (2015) Twitter sentiment classification for measuring public health concerns. Social Netw Anal Min 5(1):13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13278-015-0253-5\u003c/span\u003e\u003cspan address=\"10.1007/s13278-015-0253-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJosserand E (2007) Le pilotage des r\u0026eacute;seaux. Fondements des capacit\u0026eacute;s dynamiques de l\u0026rsquo;entreprise. Rev Fran\u0026Atilde;\u0026sect;aise de Gest 170(1):95\u0026ndash;102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3166/rfg.170.95-102\u003c/span\u003e\u003cspan address=\"10.3166/rfg.170.95-102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim G-H, Trimi S, Chung J-H (2014) Big-data applications in the government sector. Commun ACM 57(3):78\u0026ndash;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2500873\u003c/span\u003e\u003cspan address=\"10.1145/2500873\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim K-S, Sin S-CJ, Yoo-Lee EY (2014) Undergraduates\u0026rsquo; use of social media as information sources. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dr.ntu.edu.sg/handle/10356/79680\u003c/span\u003e\u003cspan address=\"https://dr.ntu.edu.sg/handle/10356/79680\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLe Moigne J-L (1994) Le Constructivisme: Tome 1: Des fondements, Tome 2: Des epistemologies. ESF\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin SY, Shanafelt TD, Asch SM (2018) Reimagining clinical documentation with artificial intelligence. Mayo Clinic Proceedings, 93(5), 563\u0026ndash;565. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.mayoclinicproceedings.org/article/S0025-6196(18)30142-3/abstract\u003c/span\u003e\u003cspan address=\"https://www.mayoclinicproceedings.org/article/S0025-6196(18)30142-3/abstract\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaciejewski M (2017) To do more, better, faster and more cheaply: Using big data in public administration. Int Rev Admin Sci 83(1suppl):120\u0026ndash;135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0020852316640058\u003c/span\u003e\u003cspan address=\"10.1177/0020852316640058\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMazzucchi N (2019) IA et industrie de d\u0026eacute;fense, le grand d\u0026eacute;fi. Revue D\u0026eacute;fense Nationale 5:86\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cairn.info/revue-defense-nationale-2019-5-page-86.htm\u003c/span\u003e\u003cspan address=\"https://www.cairn.info/revue-defense-nationale-2019-5-page-86.htm\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMehr H (2017) Artificial Intelligence for Citizen Services and Government\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMisuraca G, Van Noordt C (2020) AI Watch-Artificial Intelligence in public services: Overview of the use and impact of AI in public services in the EU. JRC Research Reports, JRC120399. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ideas.repec.org/p/ipt/iptwpa/jrc120399.html\u003c/span\u003e\u003cspan address=\"https://ideas.repec.org/p/ipt/iptwpa/jrc120399.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNutt PC, Backoff RW (1997) Organizational Transformation. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://journals.sagepub.com/doi/abs/10.1177/105649269763009\u003c/span\u003e\u003cspan address=\"https://journals.sagepub.com/doi/abs/10.1177/105649269763009\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOjo A, Raman M, Downe A (2019) Title: Toward green computing practices: A Malaysian study of green belief and attitude among Information Technology professionals. J Clean Prod 224. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclepro.2019.03.237\u003c/span\u003e\u003cspan address=\"10.1016/j.jclepro.2019.03.237\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOnder M, Saygili H (2018) YAPAY ZEK\u0026Acirc; VE KAMU Y\u0026Ouml;NETİMİNE YANSIMALARI /. ARTIFICAL INTELLEGENCE AND THE REFLECTIONS ON PUBLIC ADMINISTRATION\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParly F (2019) Avant-propos \u0026ndash; IA et d\u0026eacute;fense. Revue D\u0026eacute;fense Nationale 820(5):9\u0026ndash;17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3917/rdna.820.0009\u003c/span\u003e\u003cspan address=\"10.3917/rdna.820.0009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePencheva I, Esteve M, Mikhaylov S (2018) Big Data and AI \u0026ndash; A transformational shift for government: So, what next for research? Public Policy Adm 35:095207671878053. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0952076718780537\u003c/span\u003e\u003cspan address=\"10.1177/0952076718780537\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePollacci L, S\u0026icirc;rbu A, Giannotti F, Pedreschi D, Lucchese C, Muntean CI (2017) Sentiment Spreading: An Epidemic Model for Lexicon-Based Sentiment Analysis on Twitter. In: Esposito F, Basili R, Ferilli S (eds) Lisi (\u0026Eacute;ds.), AI*IA 2017 Advances in Artificial Intelligence, vol 10640. Springer International Publishing, pp 114\u0026ndash;127. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-319-70169-1_9\u003c/span\u003e\u003cspan address=\"10.1007/978-3-319-70169-1_9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriem RL, Butler JE (2001a) Is the Resource-Based View a Useful Perspective for Strategic Management Research? Acad Manage Rev 26(1):22\u0026ndash;40. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5465/amr.2001.4011928\u003c/span\u003e\u003cspan address=\"10.5465/amr.2001.4011928\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePurdy M, Daugherty P (2016) Why artificial intelligence is the future of growth. Remarks at AI now: the social and economic implications of artificial intelligence technologies in the near term, 1\u0026ndash;72\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun TQ, Medaglia R (2019) Mapping the challenges of Artificial Intelligence in the public sector: Evidence from public healthcare. Government Inform Q 36(2):368\u0026ndash;383. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.giq.2018.09.008\u003c/span\u003e\u003cspan address=\"10.1016/j.giq.2018.09.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSusman GI, Evered RD (1978) Scientific Merits of\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeece DJ, Pisano G, Shuen A (1997) Dynamic capabilities and strategic management. Strateg Manag J 18(7):509\u0026ndash;533. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/(SICI)1097-0266(199708)18:7%3C509::AID-SMJ882%3E3.0.CO;2-Z\u003c/span\u003e\u003cspan address=\"10.1002/(SICI)1097-0266(199708)18:7%3C509::AID-SMJ882%3E3.0.CO;2-Z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVal\u0026eacute;rie D (2019) IA, la conqu\u0026ecirc;te de l\u0026rsquo;excellence technique. Revue De?F Nationale 820(5):123\u0026ndash;130. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e\u003c/span\u003e\u003cspan address=\"http://www.cairn.info/revue-defense-nationale-2019-5-page-123.htm\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang CL, Ahmed PK (2007) Dynamic capabilities: A review and research agenda. Int J Manage reviews 9(1):31\u0026ndash;51\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWernerfelt B (1984) A resource-based view of the firm. Strateg Manag J 5(2):171\u0026ndash;180. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/smj.4250050207\u003c/span\u003e\u003cspan address=\"10.1002/smj.4250050207\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWirtz BW, Weyerer JC, Geyer C (2019) Artificial Intelligence and the Public Sector Applications and Challenges. Int J Public Adm 42(7):596\u0026ndash;615. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/01900692.2018.1498103\u003c/span\u003e\u003cspan address=\"10.1080/01900692.2018.1498103\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYami S, Lehmann-Ortega L, Naro G (2008) Capacit\u0026eacute;s dynamiques coop\u0026eacute;titives: Le cas MSI dans la sous-traitance m\u0026eacute;canique. XVIIth AIMS conference, Nice-Sophia Antipolis\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZarrad A, Jaloud A, Alsmadi I The evaluation of the public opinion-a case study: Mers-cov infection virus in ksa. 2014 IEEE/ACM 7th International Conference on Utility and Cloud, Computing (2014) 664\u0026ndash;670. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ieeexplore.ieee.org/abstract/document/7027574/\u003c/span\u003e\u003cspan address=\"https://ieeexplore.ieee.org/abstract/document/7027574/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Public Administration, Dynamic Capability, Knowledge Management, Sentiment Analysis","lastPublishedDoi":"10.21203/rs.3.rs-8290777/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8290777/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAdvances in Artificial Intelligence (AI) and its growing presence in professional settings raise questions about the transformation of public administrations, which are often addressed in speculative terms. This article seeks to demonstrate how a public administration can implement an AI initiative within its services.\u003c/p\u003e \u003cp\u003eAdopting a constructivist stance and an action research approach (Susman \u0026amp; Evered, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e1978\u003c/span\u003e), the study draws on iterative cycles between theory and field inquiry. Based on semi-structured interviews with relevant stakeholders and participant observation, the article relies on three successive qualitative studies conducted in three organizational units of the Tunisian Ministry of the Interior.\u003c/p\u003e \u003cp\u003eThe findings highlight the need to anticipate the impacts of AI tools by evaluating them in their actual context of use while actively involving the stakeholders concerned. They underscore the importance of design processes grounded in genuine needs identified by members of the organization themselves.\u003c/p\u003e \u003cp\u003eFinally, the results identify the key success factors and the challenges associated with AI integration, showing that such implementation can strengthen dynamic capabilities and knowledge management, thereby fostering a more effective adoption of technologies within public administrations.\u003c/p\u003e","manuscriptTitle":"Experimenting with an Artificial Intelligence Initiative in a Public Administration: Contributions from an Action Research Approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-22 09:25:49","doi":"10.21203/rs.3.rs-8290777/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"621a4e14-bb66-4ff7-b2cf-2ceb3fdc51ba","owner":[],"postedDate":"December 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-11T16:09:29+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-22 09:25:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8290777","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8290777","identity":"rs-8290777","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}