Living Labs for AI-Enabled Public Services: Functional Determinants, User Satisfaction, and Continued Use

Abstract

Artificial Intelligence has emerged as a transformative force in public service delivery, promising improved efficiency and responsiveness to citizens’ needs, so it is essential to understand the functional factors that influence citizens’ adoption and intention to continue using such services. Drawing on the technology acceptance model, this study investigates the influence of six functional factors, namely, usefulness, ease of use, service reliability, service quality, responsiveness, and security, on the continued use of AI-enabled public services through the mediating effect of user satisfaction. Data were collected from an online survey of AI-enabled public services in Korea during the summer of 2021, and causal mediation analysis was conducted to examine these relationships. The results indicate that usefulness, service reliability, and security significantly influenced users’ intent to continue using AI-based services. Furthermore, causal mediation analysis confirmed that the five components of AI public services—usefulness, service reliability, service quality, responsiveness, and security—had significant effects on the continued use of AI-enabled service platforms, with user satisfaction playing a mediating role in the relationships. The main functional factors can lead to higher levels of satisfaction, and this ultimately drives the sustained adoption and continued use of AI-enabled public services by citizens.

1. Introduction

Artificial intelligence (AI) is rapidly evolving with the potential to improve the usefulness of data by identifying patterns and needs internally and externally. Given its potential benefits, the use of AI in the public sector has become increasingly popular as it can enhance operations, decision-making processes, and service delivery. By leveraging AI, public sector organizations can better understand citizen needs and preferences, optimize resource allocation, and provide more personalized and efficient services to citizens. The adoption of AI technologies in government has been growing in recent years, so recent studies have examined the benefits of AI services in areas such as administrative management [1], data-driven decision-making [2,3], and public service delivery [4,5]. In particular, the COVID-19 pandemic has accelerated the adoption of AI-enabled chatbots in the public sector as a means of providing timely and accurate information to citizens regarding the virus, government regulations, and public health measures [6]. AI chatbots have played a critical role in addressing citizens’ needs during the pandemic, such as answering questions about government regulations, providing access to online services, and alleviating the burden on human resources in the public sector [7].

When it comes to AI practices and strategies, the public sector has been slower than the private sector to consider them, with less emphasis on non-technical aspects. In particular, the social, political, and contextual concerns that influence citizens’ continued use of AI-enabled public services have received limited attention beyond technical factors [8]. In the early stages of integrating AI technologies into existing government information and technology service platforms, understanding users’ experiences with AI-enabled services can help governments identify essential steps for the inclusive adoption and implementation of AI technologies for a wide range of citizens.

Although significant attention has been paid to the implementation of AI technologies in the government sector, previous research has mainly focused on conceptual aspects of AI adoption in government using qualitative methods. However, studies argued that empirical testing is much needed to improve the implications of AI use in the public sector [9,10]. Furthermore, few studies have explored the mediating effect of user satisfaction on the relationships between the explanatory factors, namely, usefulness, ease of use, service reliability, service quality, responsiveness, and security, and continued use of AI-enabled public services.

This study aims to fill this gap by empirically examining the relationships between the six explanatory variables and continued use, as well as by exploring the mediating role of user satisfaction in these relationships. This study employs causal mediation analysis using survey data on the experiences of AI-enabled public services in South Korea, which were collected from July to August, 2021. South Korea, as a technologically advanced country, has embraced the implementation of AI-based services to improve efficiency and responsiveness to citizens’ needs. Understanding the factors influencing the continued use of AI-enabled public services in Korea is crucial to ensure their successful implementation and long-term effectiveness.

This study contributes to the field of ICTs in government by examining the effects of functional factors and investigating the underlying path of the mediating role of user satisfaction in the continued use of AI-enabled public services. By exploring these relationships, this study aims to enhance the understanding of how governments can improve the adoption and utilization of the AI-based technologies and platforms. The findings of our study shed light on the importance of ICT policies, particularly in relation to adapting to the advances in the AI field. By providing evidence-based insights, our study offers valuable guidance for policymakers in implementing effective strategies that promote the sustainable progress in the evolving landscape of AI technologies to maximize their benefits for providing better public services. Furthermore, this study serves as a foundation for future research endeavors. By acknowledging the influence of cultural and contextual factors, future studies can explore cross-country comparisons to identify commonalities and differences in the adoption and impact of primary functional factors across regions.

2. Theoretical Framework

2.1. Adoption and Use of AI Services in Public Administration

AI technologies have increasingly been used in the public sector to provide better services and enhance communication with citizens. According to a recent report by the World Government Summit [11], governments worldwide are increasingly adopting AI technologies to improve service delivery, reduce costs, and increase efficiency. AI chatbots, in particular, have been widely used to provide timely and personalized responses to citizen inquiries, which can help to reduce the burden on human resources and improve citizens′ satisfaction [12]. As AI-based chatbots are becoming more sophisticated, governments are beginning to implement them in different areas, such as health, education, and public safety, to provide more comprehensive and effective services to citizens.

In recent years, AI has been increasingly implemented in public administration for improved service delivery [13]. The adoption and diffusion of AI in public administration have been extensively studied, with a focus on the factors that influence these processes [14]. It is widely recognized that the successful adoption of technological innovation is closely tied to the implementation of effective management approaches. Effective public administration approaches and practices play a crucial role in the successful implementation of new technologies, ensuring alignment with citizens′ needs and expectations and ensuring legal compliance and efficient decision-making processes [15,16]. Madan and Ashok conducted a systematic literature review, identifying technological, organizational and environmental variables and absorptive capacity as factors influencing AI adoption in public administration, and they found that perceived usefulness, perceived ease of use, perceived risk, and trust were significant predictors of mobile banking adoption among Indian consumers [13]. AI-integrated customer relationship management systems have been increasingly adopted in organizations to revolutionize customer data analysis [17]. According to Ahn and Chen, AI has been implemented in the public sector with the aim of improving the efficiency and effectiveness of service delivery [18]. By utilizing natural language processing, machine learning, and data mining technologies, AI has the potential to enhance communication between governments and citizens and provide more comprehensive and effective services to the public [19]. This study integrates existing research on AI adoption in diverse contexts and establishes a foundation for building a theoretical framework grounded in the Technology Acceptance Model (TAM) [20]. The TAM framework is particularly well-suited for examining citizens′ willingness to continue utilizing AI-enabled public services as it underscores the significance of various service characteristics, such as perceived usefulness and perceived ease of use, in shaping users′ attitudes and intentions toward technology adoption and sustained usage. Within the context of AI services in public administration, these factors are crucial for comprehending citizens′ attitudes and behaviors related to the adoption and continued use of AI services. By employing the TAM framework, this study aims to develop an in-depth understanding of the factors influencing the sustained utilization of AI services within Korea′s public sector.

2.2. Main Functional Features of AI-Enabled Platforms and the Intent of Continued Use

AI-based services can also assist citizens in navigating government websites and accessing online services more efficiently. This advantage is particularly relevant during the pandemic, as in-person visits to government offices are limited or discouraged. With the use of AI technologies, citizens can receive government services without leaving their homes, reducing the risk of virus transmission. Previous studies argued that users of information technology tended to be satisfied and continued to use services when such technologies were perceived as useful [21,22]. TAM posits that the intention to continue using technology-based services is heavily influenced by the perceived usefulness of these services, which is a key driver of technology acceptance. If users perceive these services as beneficial, valuable, and efficient, they are more likely to continue using them to accomplish their tasks more effectively. Therefore, this study proposes that perceived usefulness will have a positive influence on the continued use of AI services.

In the technological context, ease of use of a system has been considered a good predictor of IT-based services as effortless task completion is critical to the use of these services [20,23]. Some AI-based platforms or services may be challenging to use intuitively, so users who find AI public services easy to learn, understand, and interact with are more likely to keep using them since a user-friendly interface reduces barriers to continued adoption. Thus, this study hypothesizes a positive impact of ease of use on AI-based public services.

Research indicates that users perceive chatbot services to be consistently reliable, which in turn motivates their continuous usage [24,25,26]. For example, Li et al. presented empirical evidence that the reliability of chatbot services in the online travel agency market in China had a statistically significant positive effect on users’ post-use perceptions [25]. These arguments can be extended to other AI-enabled services in the public sector. Users who experience reliable AI public services with suitable functionality and quick resolution of errors are more inclined to continue using these services as they satisfy the system’s ability to deliver consistent and error-free results. Thus, this study hypothesizes a positive impact of service reliability on AI-based public services.

Some studies argue that system quality, information quality, and service quality have substantial impacts on the use of information systems and user satisfaction [27]. The provision of high-quality AI-based public services that offer up-to-date, accurate, and well-structured information has the potential to enhance user experience and foster continued use of these services. Such services are typically perceived as valuable and trustworthy sources of information, prompting users to rely on them frequently. Therefore, this study posits that service quality positively affects the continued use of AI-based services.

AI-enabled services have been found to provide personalized and responsive communication channels that can improve citizens′ perceptions of government responsiveness [6,25]. Therefore, the responsiveness component of AI-based services could be critical to the continued use of AI-based platforms. This study hypothesizes a positive impact of responsiveness on continued use. Users who receive timely updates and feedback on their inquiries, immediate action when problems arise, and prompt service provision are more likely to continue using AI public services.

The implementation of AI-based services in the public sector has also raised concerns about potential privacy and security risks [28,29]. Citizens may feel uncomfortable sharing personal information with AI chatbots, and the use of such technologies may lead to unintended consequences, such as biased or discriminatory decision-making [30]. Therefore, it is crucial to balance the benefits of AI-enabled services in the public sector with potential privacy and security concerns to expand their successful implementation. Users who trust that their personal information is securely handled and protected from potential threats are more likely to continue using AI public services as they feel confident that their privacy and sensitive information are safeguarded. Considering these discussions, this study hypothesizes a positive influence of security on users’ intention to continue using AI-enabled public services.

2.3. Mediating Paths of User Satisfaction

Citizen satisfaction and continued use of AI services are essential to generate public value [17]. The depth and breadth assimilation of AI-enabled services positively impacts citizen satisfaction, leading to the generation of public value [17]. In Korea, understanding the factors that influence the continued use of AI services is crucial to ensuring the success and sustainability of AI implementation in the public sector. Similarly, public trust and acceptance of AI services are critical factors for successful implementation in the public sector [31,32]. One study investigating the intended use of chatbots found that the initial public trust in AI chatbots in the public sector depends on the area of inquiry and the purposes communicated by the government [31]. Similarly, Gesk and Leyer conducted policy-capturing experiments to analyze AI′s acceptance in various public service scenarios, demonstrating that some specific services are still considered a human domain [32].

Previous research has demonstrated the importance of user satisfaction as a mediating factor in the adoption and continued use of technology. One study found that user satisfaction plays a crucial role in the continued use of information systems, closing the gap between initial acceptance and continued use [33]. In the context of AI services, satisfaction can be influenced by factors such as the quality of service provided, the responsiveness of the system, and the perceived security of the technology. When users are satisfied with the AI services they use, they are more likely to continue using them in the future [21,27,34]. Satisfaction can be enhanced by ensuring that the AI services are easy to use, reliable, and provide high-quality and timely responses to users′ inquiries. Moreover, maintaining a secure environment for users to interact with AI services can significantly contribute to their satisfaction and, in turn, increase their continued use intention.

Satisfied users are more likely to recommend AI services to their peers, resulting in a broader acceptance of the technology within the community. Furthermore, understanding the mediating role of satisfaction in the context of AI services can provide valuable insights for policymakers and service providers, enabling them to design and implement strategies that enhance user satisfaction and promote the continued use of AI services in the public sector. As satisfaction plays a crucial role in shaping users’ attitudes towards AI public services and determining their willingness to continue using them in the future, this study hypothesizes that user satisfaction mediates the relationships between the six explanatory variables and the continued use of AI-enabled platforms in government. By understanding and optimizing the mediating role of user satisfaction, public agencies can improve the likelihood of users adopting and consistently using AI-based public services.

Figure 1 presents the conceptual framework for examining the hypotheses proposed in this study.

3. Materials and Methods

3.1. Data

Data were drawn from an online survey conducted by a professional survey company from 13 August to 18 August 2021. The sampling frame was carefully selected to warrant equal representation of gender and individuals over 20 years of age using a stratified sampling approach.

Table 1. Demographic characteristics of respondents.

	Category	Frequency	Percentage
Gender	Male	175	50
	Female	175	50
Age	20–29	81	23.1
	30–39	88	25.1
	40–49	90	25.7
	50–59	47	19.1
	60 or order	24	6.9
Education level	High school graduate	43	12.3
	Bachelor’s degree	231	66
	Master’s degree	61	17.4
	Ph.D.	17	4.3
Job type	Public officials	19	5.43
	Teachers	17	4.86
	Professional/management	36	10.29
	Office workers	141	40.29
	Production/technology/labor	23	6.57
	Services/sales	12	3.43
	Self-employed/freelancer	26	7.43
	Housewife	31	8.86
	Students	26	7.43
	Unemployed	15	4.29
	Other	4	1.14
Annual income (USD)	None	33	9.4
	<20,000	39	11.1
	20,000–40,000	132	37.7
	40,000–60,000	73	20.9
	60,000–80,000	42	12
	80,000–100,000	18	5.1
	>100,000	13	3.7
IT competency	Level 1 (very low)	23	6.6
	Level 2 (low)	91	26
	Level 3 (medium)	140	40
	Level 4 (high)	78	22.3
	Level 5 (very high)	18	5.1
Usage of AI-based services	Rarely or never	96	27.4
	1 time per month	142	40.6
	2–3 times per month	85	24.3
	1–2 times per week	22	6.3
	Daily	5	1.4

Notes: Observations = 350; measured in Korean won (KRW) and converted to US dollars (USD) (1 USD = 1000 KRW).

presents the demographic characteristics of the respondents. The study had a balanced gender distribution, with half of the respondents being male and half female. The age groups 20–29, 30–39, and 40–49 accounted for almost 75% of the sample. A majority of the respondents held a bachelor’s degree or higher. Furthermore, most of the respondents were employed, with 40.7% identifying as office workers. In terms of annual income, 37.7% of the respondents reported earning between USD 20,000 and USD 40,000 (approximately 20 million to 40 million Korean won). Regarding information technology (IT) proficiency, more than 88% of the respondents reported being at Level 2 (26%), Level 3 (40.0%), or Level 4 (22.3%), indicating competency in IT-related services and programs. In terms of the use of public services, most participants used them less than two or three times per month (92.3%).

To ensure the relevance of the responses, all the survey participants had previously experienced or interacted with artificial intelligence-based public services, as indicated in

Table 2. Artificial intelligence-based public services in Korea.

Channel	Description
GOV.KR (https://www.gov.kr, accessed on 1 June 2021) Korean Government Portal	GOV.KR is a major government service portal that provides document issuance, government services, and policy information that each ministry offers for public use.
e-People (www.epeople.go.kr, accessed on 1 June 2021) Anti-corruption and Civil Rights Commission (ACRC)	“e-People” is a government portal of the ACRC where people can report complaints against government agencies and participate in governance to improve policies.
Call Center (110, 120,1339) Seoul City Korea Disease Control and Prevention Agency (KDCA)	The caller centers provide people and foreigners with information about government services, Seoul, and diseases or report infectious disease cases in Korea with information on living, general administrative services, and assistance with filing complaints.
People’s Secretary “Guppi” (www.ips.go.kr, accessed on 1 June 2021) Ministry of the Interior and Safety	A virtual assistant service for the public operated by the Ministry of the Interior and Safety since 29 March 2021. With this service, users can collect guides and notices from the Korean government and receive notifications on their mobile phones, use chatbots and civil service consultations, or receive guidance on civil service affairs through artificial intelligence speakers. In October 2022, the number exceeded 15 million.
AI-based call services for COVID-19 Korea Disease Control and Prevention Agency	The COVID-19 Vaccination Response Task Force launched a business with SK Telecom to implement the AI telephone guidance service, NUGU Vaccine Care Call in June 2021. Authorities at a medical institution can log into the NUGU Vaccine Care Call website and manage this AI-based telephone service for people over 75 years of age who are not familiar with smartphones. The service uses AI to remind people of their vaccination schedule by phone.
AI Monitoring Call COVID-19 Quarantine Patients Seoul City	The AI monitoring call service initiated by Seoul City was introduced in response to the need for efficient quarantine management as the number of quarantined people increased. Several local governments have used it to help reduce fatigue and reduce the workload of dedicated officials who manage quarantined people.
Beobi AI Life Legal Chatbot Service (talk.lawnorder.go.kr, accessed on 1 June 2021) Ministry of Justice	This service provides information on infectious disease notification, infectious disease reporting, epidemiological investigation, self-treatment, inpatient treatment, compensation for infectious disease patients, and support for infectious disease patients.
AI Counsellor ‘TTUBOT’ (happy.daegu.go, accessed on 1 June 2021) Daegu City	Based on the Internet and mobile web in Daegu city, Korea, Toobot provides automatic counseling to citizens 24 h a day, 365 days a year, including voice chat, complaint reporting, and inquiry for eight types of complaints.
AI service for employment placement from the Ministry of Employment and Labor (The Work AI)	The Work AI is a service that departs from the existing job search centered on occupation and automatically analyzes the job skills described in job seekers’ resumes and recruiters’ job postings and connects job seekers with the most suitable jobs.

. Out of 1790 agencies that have provided AI services in Korea, 6 were selected, and the agencies provided access to their databases to select qualifying survey participants. Out of 737 users, a total of 376 respondents completed the survey, resulting in a response rate of approximately 51%. However, 26 responses were invalid due to incompleteness and were excluded from the analysis, leaving a final dataset of 350 responses. The participants were asked to rate the level of AI-based public services based on their knowledge and experiences. All items were measured on a five-point Likert scale ranging from 1 (strongly disagree), 2 (disagree), 3 (moderate), 4 (agree), to 5 (strongly agree).

The services included information robots, mobile applications, civil services assistance, government helpline chatbots (e.g., talk.lawnorder.go.kr, happy.daegu.go.kr/dduchat, accessed on 1 June 2021), COVID-19 vaccination information, and automatic service recommendations. By encompassing all AI-enabled public services provided by public agencies in Korea, this study could provide a more comprehensive analysis of how users perceive these services and what factors motivate ongoing use, taking into account diverse user experiences.

3.2. Measurement

3.2.1. Dependent Variable

All variables in this study were evaluated and modified using measurement items previously recommended in the literature. The variable of the participants’ intention to continue using AI-based public services was evaluated using three items suggested in the work of Liu and Kim [35]: utilizing AI-enabled public services more in the future (D1), encouraging others to use AI-enabled public services (D2), and using AI public services for government-related work in the future (D3). Respondents were asked to rate their experiences on a five-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree). The internal consistency and reliability of the constructions was assessed using Cronbach’s alpha and the average inter-item covariance (AIC), as detailed in

Table 3. Reliability and descriptive statistics of measures.

Variable	Item	Kaiser–Meyer–Olkin Measure	Cronbach’s Alpha	Average Inter-Item Covariance	Mean	SD
Continued use	D4	0.718	0.856	0.354	3.617	0.643
	D5	0.710
	D6	0.769
Usefulness	A1	0.840	0.832	0.259	3.676	0.558
	A2	0.838
	A3	0.901
	A4	0.862
	A5	0.843
	B5	0.928
Ease of use	A6	0.832	0.825	0.242	3.613	0.542
	A7	0.793
	A8	0.814
	A9	0.757
	A10	0.778
Service reliability	B1	0.767	0.821	0.242	3.497	0.626
	B2	0.700
	B3	0.681
Service quality	B6	0.897	0.867	0.300	3.489	0.588
	B7	0.839
	B8	0.833
	B9	0.798
	B10	0.833
Responsiveness	C1	0.758	0.773	0.304	3.510	0.628
	C2	0.653
	C3	0.667
Security	D1	0.756	0.889	0.466	3.400	0.724
	D2	0.747
	D3	0.744
Satisfaction	C4	0.718	0.837	0.349	3.544	0.646
	C5	0.686
	C6	0.760
Channel					3.309	1.496
Gender					0.5	0.501
Age					0.517	0.5

Note: Observations = 350.

. The Cronbach’s alpha coefficient of 0.856 signified the reliability of the items in constructing the measure for continued use. Additionally, the AIC score of 0.354 demonstrated the discriminant validity of the items employed in assessing the continued use of AI-enabled public services. Furthermore, the conceptual suitability of these items was validated by the Kaiser–Meyer–Olkin (KMO) test, which indicated scores of above 0.71 for all three items.

3.2.2. Explanatory Variables

This study adopted the perceived usefulness and ease of use constructs from the work of Gefen and Straub [36]. Six items were used to measure the usefulness variable: I can complete tasks faster using AI public services (A1), AI-enabled public services will be a valuable service for me (A2), the content of AI public services will not be beneficial to me (A3), AI public services will improve the efficiency of service search and utilization (A4), AI public services are useful (A5), and I can easily search for related web pages using AI public services (B5). The item regarding the perceived lack of usefulness of the content of AI public services will not be beneficial to me was coded in reverse. As evidenced by a Cronbach’s alpha of 0.832, an AIC score of 0.259, and KMO scores ranging between 0.84 and 0.93, the six items were found to be reliable and consistent in measuring the intended construction of the usefulness variable. The ease of use variable was measured using five items: learning to interact with AI public services will be easy (A6), interacting with AI public services will be clear and understandable (A7), I will have no problem interacting with AI public services (A8), I can use AI public services proficiently (A9), and AI public services will not be difficult to use (A10). The reliability of these five items in measuring ease of use was confirmed by a Cronbach’s alpha of 0.825, an AIC score of 0.242, and KMO scores ranging between 0.76 and 0.83.

The measurement items for service reliability, service quality, and responsiveness were taken from previous studies [28,37] and were modified to be more appropriate for the context of AI-based public services. Service reliability was assessed using three items, namely, the reliability of the AI public services operation (B1), the ability to quickly resolve errors that occur while using the AI public services (B2), and the proper functioning of all the functions and services offered by the AI public services (B3). A Cronbach’s alpha of 0.821, an AIC score of 0.242, and KMO scores greater than 0.68 validated the reliability of these three items in constructing the service reliability variable. Service quality was evaluated using five items: AI public services are arranged in a user-friendly manner (B6), AI public services provide the latest updated information (B7), AI public services are designed to find the necessary information quickly (B8), AI public services provide well-organized information (B9), and AI public services provide accurate information on essential services (B10). A Cronbach’s alpha of 0.867, an AIC score of 0.3, and KMO scores exceeding 0.80 indicated that these five items were reliable in measuring service quality.

Responsiveness was constructed by three items: informing users of the progress of their complaints via email or messages (C1), taking immediate action when problems occur (C2), and timely provision of AI public services (C3). The reliability of these three items to construct responsiveness was confirmed by a Cronbach’s alpha of 0.773, an AIC score of 0.304, and KMO scores ranging between 0.65 and 0.76. Security was evaluated using three items: the safe handling of personal information provided when using the AI public services system (D1), implementation of security measures to prevent the leakage or theft of shared personal information (D2), and effective protection of personal information provided to the AI public services system from potential hacker attacks (D3). A Cronbach’s alpha of 0.889, an AIC score of 0.466, and KMO scores exceeding 0.75 validated the reliability of the three items in the construction of security.

3.2.3. Mediating Variable

User satisfaction with AI public services was measured using a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree), based on three items suggested by previous studies [34,38]. The items are as follows: I am satisfied with general AI public services (C4), I am very satisfied with my experiences using AI public services (C5), and AI public services provide services that meet my needs (C6). The reliability and consistency of the three items in constructing user satisfaction were confirmed by a Cronbach’s alpha of 0.837, an AIC score of 0.349, and KMO scores ranging between 0.69 and 0.76.

3.2.4. Control Variables

This study included three control variables: the channel of AI public services, gender, and age. In line with the efforts of the Korean government to promote the development and adoption of AI in government services, nine official AI service platforms are available to access these services, as detailed in Table 1.

Table 4. Frequency of control variables.

Variable	Code	Frequency	Percent
Channel	Gov24	19	5.43
	Sinmungo (e-people)	83	23.71
	Call centers	123	35.14
	Secretary Guppi	74	21.14
	COVID-19 vaccination AI call service	30	8.57
	AI call service for COVID-19 self-quarantine	6	1.71
	Bubby AI life legal knowledge service	5	1.43
	AI counselor TuBot	3	0.86
	Work AI	7	2
Gender	Female (0)	175	50
	Male (1)	175	50
Age	20–30 (0)	169	48.29
	Over 40 (1)	181	51.71

, which presents the frequency of participants accessing AI-enabled public services, indicates that the call center platform was the most frequently accessed among the participants, with approximately 35% of users. The Sinmungo platform ranked second with about 24% of users, and the Secretary Guppi platform followed closely behind with about 21% of users. Notably, participants mostly used these three channels, which together accounted for approximately 80% of usage.

Gender was coded as 0 for female participants and 1 for male participants. Regarding age, individuals between the ages of 20 and 30 were coded as 0, representing approximately 48% of the participants, while individuals over the age of 40 were coded as 1, representing about 52% of the participants. Both gender and age categories were evenly distributed across all groups to ensure a balanced representation of participants in this study.

3.3. Analytic Analysis

This study used causal mediation regression analysis to examine the mediating effect of satisfaction on the relationship between the explanatory variables and the dependent variable. As the mediating variable partially or fully explains these relationships, traditional mediation analysis involves a series of regression analyses to estimate the direct and indirect effects of the independent variables on the dependent variable while controlling for the effect of the mediating variable [39]. The first step of regression analysis tests the effects of the independent variables on the dependent variable. The second step of the regression analysis involves testing the correlation between the independent variables and the mediating variable. The third step of regression analysis is a full model that includes the independent variables, the mediator, and the dependent variable.

However, there are concerns about the traditional mediation approach in relation to its statistical limitations and the challenges in identifying causal mediation effects [40]. Recent studies have proposed a causal mediation technique to explain the mediator’s degree of the mediating effect on the interaction between the independent variables and the dependent variable [40,41]. This technique decomposes the total effect of the independent variable into two components: direct and indirect portions. The direct effect represents the effect of the independent variable on the dependent variable that is not transmitted through the mediator, while the indirect effect represents the effect of the independent variable on the dependent variable that is transmitted through the mediator.

4. Results

The results of three regression analysis models are presented in

Table 5. Regression analysis results.

	Model 1 Continued Use	Model 2 Satisfaction	Model 3 Continued Use
Usefulness	0.403 *** (0.0580)	0.207 *** (0.0552)	0.343 *** (0.0570)
Ease of use	0.0233 (0.0571)	0.0430 (0.0543)	0.0108 (0.0550)
Service reliability	0.159 *** (0.0552)	0.186 *** (0.0525)	0.105 * (0.0541)
Service quality	0.0991 (0.0628)	0.282 *** (0.0597)	0.0174 (0.0624)
Responsiveness	−0.0122 (0.0479)	0.235 *** (0.0455)	−0.0803 * (0.0479)
Security	0.321 *** (0.0393)	0.109 *** (0.0374)	0.290 *** (0.0383)
Satisfaction			0.290 *** (0.0549)
Channel	0.0253 * (0.0140)	−0.0280 ** (0.0134)	0.0334 ** (0.0136)
Gender	−0.0546 (0.0418)	0.00811 (0.0397)	−0.0570 (0.0402)
Age	−0.0308 (0.0419)	−0.108 *** (0.0399)	0.000667 (0.0408)
Constant	0.0600 (0.161)	−0.0583 (0.153)	0.0769 (0.155)
Observations	350	350	350
R-squared	0.663	0.697	0.689

Notes: *** p < 0.01, ** p < 0.05, * p < 0.1; standard errors shown in parentheses.

. In Model 1, the six independent variables are regressed against continued use without involving the mediator of user satisfaction. The coefficient values of the functional variables indicate their total effects on the continued use of AI-enabled public services. As hypothesized, the three variables—usefulness, service reliability, and security—were statistically significant predictors of the continued use of AI-enabled public services at the 99% confidence level. Of these factors, usefulness was identified as the most significant predictor of encouraging users to continue using AI-based public services. As experienced users have tended to be primarily concerned about privacy and security of their personal information when using internet services [28,29], the security-related challenge was prevalent in the case of AI-based services in this study. The results support that users’ confidence in the reliability of AI-enabled systems plays a key role in shaping their intention to continue using AI-based public services.

However, the other three variables, namely, ease of use, service quality, and responsiveness, had no statistically significant influence on continued use. The insignificant finding for ease of use on continued use differs from the results of Ashfaq et al., whose study identified ease of use as a significant predictor of continued use of a chatbot e-service [21]. This discrepancy can be attributed to various factors, such as the use of different AI-based service channels, the ways in which AI-based services are accessed, or the intended purposes for which AI-based services are used. This rationale could also explain the insignificant effect of service quality on the continued use of AI services. Users may perceive AI-based public services as an additional feature of existing information and communication technologies services, leading them to prioritize service quality less than usefulness, service reliability, and security. Regarding responsiveness, it was unexpectedly negatively associated with continued use, although it was not statistically significant.

Model 2 entailed regressing the six functional variables of user satisfaction. Although this study did not formulate specific hypotheses to test relationships between these variables and user satisfaction, it is worthwhile to note their associations in order to observe the mechanisms underlying their effects on user satisfaction. The results indicate that five independent variables significantly influenced user satisfaction at the 99% confidence level. These findings confirm recent arguments that the usefulness of AI-enabled services [21,27,34], service reliability [21], service quality [2,42], responsiveness [43], and security [43] could have positive impacts on user satisfaction. However, ease of use did not have a statistically significant impact on user satisfaction, consistent with the findings of Model 1. In a related study, Ashfaq et al. proposed a positive correlation between ease of use in the context of online shopping and user satisfaction, but this relationship was deemed insignificant [44]. This may imply that users who have previously used AI-enabled public services have a certain level of technological proficiency, rendering ease of use a less salient factor in determining user satisfaction.

In Model 3, both the independent variables and the moderator of satisfaction regressed on continued use. The coefficient values of the independent variables showed their direct effects on the continued use of AI-enabled services through the mediator’s pathway.

In this study, we controlled for the impact of the AI technology platform and found a significant effect of the AI channel on the continued use of AI-based services, as presented in Models 1 and 3. The Korean government has explicitly adopted AI technologies to interact with citizens, including chatbots, virtual assistants, and automated customer service systems that allow users to perform search and scheduling using reasoning and predictive patterns. This finding implies that users may have varying experiences when using different types of AI-based public services, which could potentially impact their willingness to continue using AI-based services in general. Regarding gender, the results from Models 1 and 3 indicated that gender was not significantly associated with the continued use of AI-based services. However, a negative direction was observed, indicating that male users were less inclined to continue using AI-based services. As for age, the result was statistically insignificant, but its negative association shown in Model 1 indicated that participants over 40 years old were less likely to continue using AI-based services than participants in their 20s and 30s.

Table 6. Mediation analysis results: indirect effects of the mediator.

	Estimate	SD	p > z
Usefulness	0.060 **	0.020	0.002
Ease of Use	0.012	0.016	0.433
Service Reliability	0.054 **	0.018	0.003
Service Quality	0.082 ***	0.023	0.000
Responsiveness	0.068 ***	0.018	0.000
Security	0.032 **	0.012	0.011

Notes: *** p < 0.01, ** p < 0.05; standard errors shown in parentheses; mediation estimate (natural indirect effect) = Model 1 (total effect) − Model 3 (controlled direct effect).

presents the results of the causal mediation analysis conducted to assess the mediating effect of user satisfaction in the relationships between the six explanatory factors and continued use. The findings reflect the indirect effects of user satisfaction, which significantly mediate the influence of the five independent variables, namely, usefulness, service reliability, service quality, responsiveness, and security, on continued use at the 95% confidence level. As hypothesized, each of the five independent variables was positively mediated by user satisfaction to enhance its impact on continued use. Among these factors, the indirect effect of user satisfaction was high on the relationship between service quality and continued use. While user satisfaction played a notable role, the impact of service quality on continued use, mediated by user satisfaction, was not found to be statistically significant. In the case of responsiveness, the initially insignificant total effect became significant when its controlled direct effect was examined at the 90% confidence level. Therefore, this finding supports the assertion that user satisfaction helps explain some of the variability in the continued use of AI-based services, allowing the effect of responsiveness to become significant.

However, the findings indicate that user satisfaction does not serve as a significant mediator between ease of use and continued use of AI-enabled public services. This implies that even though users are satisfied with these services, it does not considerably increase the explanatory power of ease of use for the perceived continued use. This result could be attributed to the lack of a statistically significant impact of ease of use, as shown in Table 3. Moreover, the relatively small mediating effect observed may not have sufficient power to make the total effect of ease of use statistically significant.

5. Discussion

This study intends to broaden our understanding of the main functional elements of AI-enabled public services that impact users’ continued use of AI platforms, as well as the mediating role of user satisfaction on these associations. Previous research has argued that the value of AI technologies in government is primarily related to their ability to improve efficiency, performance, economic outcomes, service delivery, decision-making, and interaction with citizens [10]. Taking into account the desired outcomes of the implementation of AI technologies in government, this study examined whether the principal functional components of AI-enabled platforms remain significant predictors for the continued use of AI-based public services, similar to other technologies and web-based tools. Moreover, this study expands the scope of research to identify a specific pathway underlying the relationships between the six factors and continued use. The mediating mechanism may provide valuable information to develop more effective strategies that enhance the mediating effect, in addition to improving the functional determinants of the continued use of AI-based services.

The discussions in this study to investigate citizens’ intentions to continue utilizing AI technologies in government are primarily grounded in TAM. Recent studies have developed an integrated model that incorporates the expectation-confirmation model, theory of planned behavior, TAM, innovation diffusion theory, united theory of acceptance and use of technology, and artificial neural networks to predict users’ continued intention to use technological tools in the educational context [45,46,47,48]. Given that this study proposes the critical mediating role of user satisfaction, adopting the expectation-confirmation approach could provide additional explanation about the impact of perceived usefulness and satisfaction on the continued use of AI applications in government. For instance, in their study, Al-Sharafi et al. [46] asserted that the perceived usefulness of chatbots and satisfaction positively impacted learning outcomes when the performance experiences of chatbots met users’ initial expectations. Incorporating an integrated approach into future research efforts will allow us to broaden the theoretical perspectives surrounding the adoption and continued use of AI technologies in the public sector.

We confirmed that usefulness, service reliability, and security significantly impacted users’ intent to continue using AI-based services, which aligns with the technology acceptance model. In contrast, ease of use, service quality, and responsiveness had no statistically significant influence on the continued use of AI applications. This discrepancy in the results from those previous studies may be attributed to various factors, such as different AI-based service channels, ways of accessing AI-based services, or their intended purposes. Although this study controlled for the effects of AI-enabled service platforms used in the public sector, different AI-based service experiences with different platforms could result in varying intentions to continue using these services, as acknowledged in our findings. Future research could explore the effect of different AI-based platforms (e.g., autonomous agents, tracking and monitoring practices, virtual assistants, citizen-centric applications) on users’ intent to continue using them, thereby offering practical implications for the effective adoption of AI technologies in government.

The study also highlights the mediating role of user satisfaction in the relationships between the six functional variables and the continued use of AI-enabled public services. User satisfaction significantly mediated the relationship between the five variables (usefulness, service reliability, service quality, responsiveness, and security) and continued use, enhancing the impact of these variables on users’ intentions of using AI platforms for public services. However, user satisfaction did not serve as a significant mediator between ease of use and continued use. Although this result implies users are satisfied with these services, it does not considerably increase the explanatory power of ease of use for the perceived continued use. This may be due to the users’ relatively high technological proficiency, making ease of use a less important factor in their satisfaction and continued use intentions. Fostering user satisfaction with AI service platforms is imperative to advancing the implementation of AI technologies in government. The results of causal mediation analysis in this study imply that in the design and implementation of AI-enabled service platforms, improving user satisfaction should be a crucial consideration.

Some limitations are recognized related to data and measurement issues. This study primarily focused on the impact of six functional factors on the continued use of AI-enabled public services. However, it is recognized that other factors, such as trust in government and privacy, can also serve as significant predictors in this context. The survey in this study did not include specific questions addressing these additional explanatory factors due to the need to limit the survey length and minimize its complexity to ensure a high response rate. Future research may consider examining the impacts of trust in government and privacy-related matters on the continued use of AI-enabled public services.

While the questionnaire was designed based on previous studies [28,34,36,37,38] and the reliability and consistency of the survey items were validated, common method bias and response bias may still pose concerns for this cross-sectional online survey. With regard to responsiveness, its total effect and controlled direct effect were found to have a negative effect on the continued use of AI-based services, although these effects were not statistically significant. This negative direction may be attributed to response bias. Users may perceive overly responsive AI services as intrusive, leading to a reluctance to continue using them. Additionally, since AI-enabled public services are relatively new to citizens, the overly responsive services may be perceived as robotic or impersonal, making users feel less connected to such services and reducing their motivation to continue using them. Future research could employ a mixed methods approach to gain a deeper understanding of user experiences associated with personalized AI-based services in government. This approach would enable researchers to collect qualitative data to offer more comprehensive insights into user experiences and their perceptions of AI-enabled services in the government context.

6. Conclusions

This study provides empirical evidence to the growing literature on user perceptions of continued use of AI-enabled public services, specifically in relation to the variables of usefulness, ease of use, service reliability, service quality, responsiveness, security, and satisfaction. Notably, since user satisfaction serves as a critical factor in linking the functional attributes of AI-based platforms or services to user intentions to continue using them, this study asserts the significant mediating role of user satisfaction in the context of AI-enabled government services, with the exception of the ease of use attribute. Given that usefulness and security are the most significant factors in promoting sustained use of AI-enabled public services through the user satisfaction, public agencies should prioritize the development and implementation of AI innovations that are both highly efficient and meet the highest standards of security. This strategy can build trust with users, resulting in higher user satisfaction and encouraging the continued use of these services.

This study reveals the importance of enhancing functional factors—usefulness, service reliability, and security—to increase users’ adoption and intention to continue using AI-enabled public services. Zhang et al. emphasized the importance of aligning AI-enabled services with citizens’ needs and preferences, suggesting that user-centered design approaches can be used to understand user requirements and tailor services accordingly [49]. Beyer and Holtzblatt argued that establishing robust IT infrastructure and monitoring systems is crucial for ensuring service reliability and prompt issue resolution [50]. Additionally, stringent data protection measures, including encryption and transparent communication about data usage, as highlighted by Zhang et al. [49], can enhance security and build trust among users. By adopting these strategies, policymakers can effectively improve the identified functional factors, leading to enhanced adoption and continued use of AI-enabled public services.