鞠京芮、孟庆国等：Citizen preferences and government chatbot social characteristics: Evidence from a discrete choice experiment
Abstract
Government chatbots have become increasingly popular as artificial-intelligence-based tools to improve communication between the government and its citizens. This study explores the interaction mode design of a trustworthy government chatbot, which involves multiple social characteristics from the user-centric perspective. A discrete choice experiment was conducted in the context of Chinese government chatbots to examine the effects of various social characteristics on citizen preferences. Participants utilized a crowdsourcing survey platform to report their preferences for interaction processes designed with distinct sets of social characteristics. Valid data were obtained from 371 participants and analyzed using a multinomial logit model. The results indicate that (in order from highest to lowest impact) emotional intelligence, proactivity, identity consistency, and conscientiousness significantly influence the citizens' preferences. Identity consistency has a negative effect, whereas the other factors all have positive impacts. It was also determined that some of these correlations are influenced by the participants' individual characteristics, such as age, gender, and prior experience with chatbots. This work provides empirical evidence for the relative importance of social characteristics and their impacts on user perception, expands the service dimension scope of information provision/communication (one of five categories of digital interaction), and facilitates the identification and operationalization of the social characteristics. We provide a theoretical framework to understand the interaction model design of a trustworthy government chatbot and also offer practical recommendations for government chatbot designers and policy implications.
Introduction
In recent years, many countries have employed artificial intelligence (AI) technologies to transform their digital government services. Mechanisms have been established at the national level (e.g., agencies, projects, pilot initiatives) to fully exploit the potential of AI for policy making and the design of government service programs. The 2020 United Nations E-Government Survey reported that the number of countries that use chatbots (i.e., AI-enabled user-interaction applications) in their national portals doubled from 28 in 2018 to 59 in 2020. Government chatbots act as virtual civil servants that are available around the clock. They use AI-related algorithms, such as natural language processing (NLP), deep learning, knowledge graphs, and decision trees, to analyze citizen inquiries and respond instantly and accurately. They also have the capability of continually “learning” about citizens' needs in order to optimize their responses. This study focuses on text-based chatbots that provide consultations regarding government services, rather than voice-based or physical bots.
With the advent of new chatbots such as GPT3 (invented by OpenAI), it is argued that chatbots are competent to change the world, as they are capable to chat exactly like humans. However, despite the impeccable communication abilities, trust still remains a critical issue with these advanced chatbots (New York Times, 2022). Further, industry reports suggest that there are already a billion users of text or voice-based conversational chatbots (Singh, 2021). Hence, it is likely that over time, chatbots will be universally adopted by multiple people to interact with the government, thus it is vital to understand the interaction model design of a trustworthy government chatbot. This is the focus of our study.
By the end of November 2019, about 70% of the provincial governments in China had launched chatbots on their portals. These chatbots exhibited varying levels of performance in the interaction model design (see more in Table 3). For example, some chatbots use an official language style (e.g., response to citizen: Sorry, I can't find the results for your query.), while others use an unofficial language style (e.g., response to citizen: Your question is so difficult that I can't answer it. I've written it down and tell you in a few days!). When dealing with a relatively complex inquiry, some chatbots use multi-turn dialogue to gradually ascertain a citizen's needs, whereas others use single-turn dialogue to solve the query all at once. Therefore, the pertinent question becomes: which design can better help governments interact with citizens? This issue has clearly not received enough attention from the government. Instead, public administrations usually focus on technologies and existing processes during public e-service development, rather than thinking of the preferences of the end-users, namely citizens (Rose, Flak, & Sæbø, 2018). The resulting e-services, which ignore user preferences, often lead to low adoption rates (Fakhoury & Aubert, 2015) and low service quality evaluations (Buckley, 2003). Actually, individual preferences should be central to research regarding e-service success factors (Wirtz, Weyerer, & Rösch, 2019). A user-centric approach, resulting in the development of e-services from an end-user's perspective, must be sufficiently explored (Högström, Davoudi, Löfgren, & Johnson, 2016). Specifically, governments should evaluate citizens' preferences in order to ensure that citizens receive the services provided by the government chatbots more effectively, thereby engendering a sense of satisfaction toward their government (Lin & Doong, 2018).
However, research into user-centric service design in the context of e-government is still in its infancy. Based on the five categories of digital interactions proposed by Jansen and Ølnes (2016), Pleger, Mertes, Rey, and Brüesch (2020) identified seven possible characteristics of public service, including registration, infrastructure, communication, data security/data protection, processing status, time expenditure, and price. Further, they tested user preferences for these aspects through a public survey with conjoint analysis. Among the five categories of digital interactions, the interaction between the chatbot and a citizen evaluated in our study represents a type of information provision/communication. While, Pleger et al. (2020) identified only one general service characteristic for this interaction — communication — which alone cannot fully encompass the characteristics of the complex human-like interactions between a government chatbot and a citizen. Moreover, the chatbots are capable to exhibit eleven possible social characteristics that benefit human-machine interactions, such as proactivity, conscientiousness, communicability, damage control, thoroughness, manners, moral agency, emotional intelligence, personalization, identity, and personality Chaves & Gerosa, 2021). Therefore, we aim to address two key research questions intended to shed light on citizens' preferences in terms of the social characteristics of a government chatbot: (i) what are the crucial social characteristics of a government chatbot, and (ii) how do they affect a citizen's preferences for the interaction?
The remainder of this paper is organized as follows. Section 2 reviews the existing literature on the social characteristics of government chatbots and citizen preferences. 3 Hypotheses, 4 Method, 5 Results, respectively, describe the hypotheses informed by the presented literature review, present the details of the applied research method, and report the empirical findings of this study. Section 6 discusses the theoretical and practical implications, as well as the limitations of this study. Section 7 presents the conclusions drawn based on the results discussed in this article.
Section snippets
Digital interactions between a government chatbot and a citizen
Government-to-citizen (G2C) e-government refers to government systems using information and communication technology (ICT) to better serve their citizens. It aims to simplify and improve transactions, improve public service delivery, and provide benefits to end-users (Al-Hujran, Al-Debei, Chatfield, & Migdadi, 2015; Axelsson, Melin, & Lindgren, 2013; Moon, 2002). Recently, there has been growing interest regarding the potential of government-focused digital solutions (Buckley, 2003;...
Hypotheses
This study systematically identifies five key social characteristics of government chatbots, which enable government chatbot design to better meet practical requirements (Porreca et al., 2018; Tavanapour et al., 2019). These key features also encompass the three necessary dimensions of the conceptual model of chatbot social characteristics (Chaves & Gerosa, 2021), thus highlighting the uniqueness of chatbots in digital interactions (compared with other e-government services). Based on previous...
Discrete choice experiment
The discrete choice experiment (DCE) is a widely-used research method in the marketing, public administration, and information systems fields (Cantarelli, Belle, & Longo, 2020; Jensen & Pedersen, 2017; Richard, Coltman, & Keating, 2012; Van Puyvelde, Caers, Du Bois, & Jegers, 2016). The DCE method was employed in this study because it allows decisions made by subjects to resemble their real-world decision-making process more closely than in other methods of evaluating individual preferences (...
Results
The survey sample comprised a broad range of participants. As shown in Table 4, 93% of the participants were between 18 and 40 years old, 47% were male, 74% had bachelor's degrees, 11% had graduate degrees, and 15% did not have any university degree. The majority of our sample participants are young and have a certain level of education. Because this population is more likely to become the early adopters of chatbots (Jain et al., 2018; Kasilingam, 2020), we believe that it is reasonable to...
Discussion
We hypothesized that proactivity embedded in the government chatbot would positively impact the citizens' preferences for interacting with the chatbot (H1). As shown in Table 5, the impact of this social characteristic is significant; therefore, H1 is supported by the experimental data. Our theoretical explanation for the significance of this impact is that a government chatbot with high-level proactivity (i.e., one that provides additional information) is more successful in attracting citizens...
Conclusions
We are entering in an era where it is increasingly becoming difficult to identify chatbots from humans. Thus, it is crucial to understand the interaction model design of a trustworthy government chatbot for its enhanced efficacy. In the current study, we investigate the key social characteristics of government chatbots and understand how they impact citizen preferences., We have identified five important social characteristics (i.e., proactivity, conscientiousness, communicability, emotional...
Acknowledgements
This research was supported by the National Natural Science Foundation of China [72004110, 72034001, 71974044, 71974111], the China Postdoctoral Science Foundation [2020M680025], the Heilongjiang Provincial Natural Science Foundation of China [YQ2020G004] and the Fundamental Research Funds for the Central Universities [HIT.OCEF.2022054 and HIT.HSS.DZ201905]...
