This study aims to identify the key predictors that affect the continuance intention of artificial intelligence personal assistant [AIPA]. It proposes the theoretical framework which employs utilitarian value, hedonic value, perceived ease of use, perceived usefulness, novelty value, perceived enjoyment, and parasocial interaction. Data was collected from 257 users of artificial intelligence personal assistants. This study used partial least squares structural equation modeling [PLS-SEM] to analyze the empirical data. The results show that utilitarian value and hedonic value are significantly correlated with continuance intention. The findings reveal that perceived ease of use, perceived usefulness, novelty value have a significant effect on utilitarian value. The analysis results indicate that novelty value, perceived enjoyment, and parasocial interaction are significantly associated with hedonic value. The current study conducted multi-group analysis according to the AIPA type, gender, and use experience. The results of this study will be a useful guideline for research and business on AIPA.
Keywords: Artificial intelligence personal assistant, Human-computer interaction, Continuance interaction, Utilitarian value, Hedonic value, Novelty value
Artificial intelligence personal assistant; Human-computer interaction; Continuance interaction; Utilitarian value; Hedonic value; Novelty value.
1. Introduction
With the development of emerging devices such as smartphones, tablets, and smart TVs, artificial intelligence personal assistant [AIPA] is also evolving. AIPA uses natural language in processing to answer questions, make recommendations, and activate mobile apps []. It usually recognizes the user's voice and processes the information []. Users can start phone calls, send emails, and find locations on a map via AIPA []. In addition, Google Duplex can now book hair salons or make restaurant inquiries over the phone []. AIPA market is expected to be worth $50.9 Billion, Globally, by 2028 at a 30% CAGR []. Apple's Siri continues to claim the largest relative market share among smartphone assistants. Its share became low slightly in 2018 but still came in at 45.1% compared to 29.9% for Google Assistant, 18.3% for Amazon Alexa, and 4.7% for Samsung Bixby []. Against this background, it would be valuable both academically and practically to study AIPA users' continuance intention. Therefore, this research identifies the key factors that influence continuance intention.
Users can get utilitarian and hedonic benefits from AIPA []. The utilitarian value of the information system [IS] increases the continuance intention []. Users are more likely to use voice personal assistants when assistants provide more useful benefits []. The hedonic value of information technology [IT] significantly determines behavioral intentions such as adoption intention and continuance intention []. The subjects of this study are users of AIPA installed on smartphones. Users can easily access AIPA because they always carry their smartphones. They often use AIPA to soothe their boredom or to find fun. If AIPA provides users with greater hedonic value, users' continuous intentions may increase. Based on the above results and discussion, this study assumes that utilitarian value and hedonic value are preeminent antecedents of AIPA users' continuance intention.
Perceived ease of use and perceived usefulness are predictive variables of behavioral intention in the context of IT [; ; ; ]. Perceived usefulness has a significant effect directly or indirectly on AI users' continuance intention [, ]. As AIPA is easier to use and provides more useful information to users, users would get greater utilitarian value from their assistants. The novelty value is related to originality and uniqueness []. Compared to decision support technologies before AIPA, AIPA offers new experiences in the following respects. First, AIPA communicates like humans and recognizes natural language accurately []. Existing digital assistants have processed information using text or computer code. AIPA receives command through voice and processes information directly. Second, AIPA operates multiple functions of smartphones on its own []. Third, AIPA also provides friendly conversations to users and serves as a secretary close to humans [; ]. Lastly, AIPA can be customized for users []. This is a completely new experience that was not seen before the emergence of advanced AIPA. This novelty may shape utilitarian benefits. Therefore, this study postulates that perceived ease of use, perceived usefulness, and novelty value play an essential role in generating utilitarian value.
Perceived enjoyment is the main explanatory variable of the hedonic IS []. If users experience joy and fun through communication with AIPA, they would feel higher level of hedonic values from AIPA. Parasocial interaction refers to a kind of psychological relationship that users experience during conversations with AIPA []. The user may engage in a parasocial interaction with AIPA to request counseling or talk about his/her mind []. In particular, as the time to stay at home to prevent COVID-19 has increased recently, users have interacted with AIPA to mitigate mental damage []. The greater these parasocial interactions are formed, the higher the user's hedonic value may be. AIPA also responds to user jokes and chatter. These new reactions would generate hedonic value. Therefore, this study suggests that perceived enjoyment, parasocial interaction, and novelty value would drive hedonic value.
The current study aims to figure out the predictors of continuance intention of general AIPA users including business and academic areas. Even if users are engaged in various fields, there may be common factors for the continuance intention of AIPA. Moreover, this study choose the factors influencing continuance intention based on the following rationale. First, utilitarian value and hedonic value were set as the main preceding factors because AIPA provides both usefulness and fun. AIPA can easily schedule and offer interesting search results. Utilitarian value may be formed by useful function and ease of use, and hedonic value may be determined by enjoyment and parasocial interaction. Since AIPA can be new in terms of functionality or enjoyment, novelty value would serve as the predominant antecedent of both utilitarian value and hedonic value. Second, this work endeavored to suggest new factors and paths that have not been investigated in the literature on AIPA users. Previous studies have employed an IS success model [], a technology acceptance model [TAM] [], an expectation confirmation model [ECM] [], a unified theory of acceptance and use of technology [UTAUT] [], a technology failure factors [], and privacy variables [; ]. The above research has not validated the paths that this study develops. Last, this research focused on factors specialized in the nature of AIPA. AIPA offers services through interaction with users. Users may build a parasocial interaction with AIPA because it also provides a helpful dialogue for human emotions. According to the above logic, thepresent study adopted ease, usefulness, uniqueness, pleasure, and interaction.
The structure of this study is as follows. Section presents the background and related works. Section describes the research model and hypotheses. Section details the methodology. Section provides the results of analysis. Section covers the discussion. Finally, Section shows implications, limitations, and future research directions.
2. Background and related work
This study searched the online DB of Web of Science, Google Scalar, and leading publishers [e.g., Elsevier, Emerald, Springer, Taylor and Francis, and John Wiley & Sons] to review the relevant literature. The keywords for the research subject were AIPA, artificial intelligence assistant, intelligent virtual assistant, digital assistant, and voice assistant. To intensively review research on user behavior, this work also reflected keywords such as behavioral intention, continuance intension, influencing factor, and affecting factor. Numerous studies have analyzed the behavior of users of AIPA [ ; ]. Several works have identified the main factors of AI users’ behaviors from the perspective of an IS [, ]. Some research empirically verified the role of interaction based on the fact that AI interacts with humans [, ]. Prior studies have mainly focused on smartphone artificial assistants, in-home speakers, and other smart device assistants [, , ]. From the next paragraph, this article introduces previous studies reflecting utility and pleasure, former research introducing IS factors, and related studies based on a marketing perspective.
Some studies employed utility and pleasure in explaining the behavior of personal assistant users and analyzed them. examined the drivers of users’ continuance usage intention of smart voice assistants. They provided evident support that trust, satisfaction, slowness of adoption, skepticism, user engagement, and attitude are the significant enablers of continuance usage intention in the voice assistant context. Particularly, the attitude was the second-order construct that consisted of a utilitarian attitude and a hedonic attitude. According to the results, users who have a utilitarian attitude and a hedonic attitude are more likely to use AIPA continuously. explored the factors affecting the usage of an In-home voice assistant. They indicated that utilitarian benefits have a significant effect on usage. It was found that hedonic benefits do not affect usage. In-home voice assistant usually uses the same software as the assistant on the smartphone. Both Amazon Alex and Google Assistant run on smartphones as well as on smart speakers [e.g., Echo Dot and Google Nest] or other devices [e.g., Echo Show or Google Home]. Since AIPA on a smartphone allows conversations anytime and anywhere, it may provide greater fun to users than in-home assistants. Therefore, this study reflects two components which are utilitarian value and hedonic value. Although Both and employed a utilitarian perspective and hedonic perspective, they did not account for the preceding factors of utility and pleasure. The present research observes the formation process of utilitarian value and hedonic value.
A battery of studies has analyzed AI users' behavior by reflecting major factors that have been verified in the literature on ISs [, , ]. proposed the theoretical framework to account for the continuance intention of AI chatbots. They incorporated the constructs in DeLone and McLean's IS success model and variables in ECM. It was unveiled that trust, user satisfaction, and perceived usefulness are significantly associated with continuance intention. Also, information quality and service quality had a positive relation to user satisfaction. The ECM describes the continuing behavior of users of the ISs []. According to the model, confirmation drives perceived usefulness and satisfaction. Satisfaction forms the continuance intention. The ECM was applied on AIPA because confirmation, perceived usefulness, and satisfaction can lead to continuance intention in the context of AIPA. By considering the results in , this study posits perceived usefulness as an exogenous variable within the research model. reflected essential factors which had been extensively demonstrated in IT contexts. The current study is different in that it adds AIPA-specific variables while maintaining the core determinants of IT use. developed a research model by combining the factors in IS success model and components of TAM to clarify the key factors influencing the continuance intention of the voice user interface. They discovered that perceived usefulness, perceived enjoyment, mobile self-efficacy, and trust significantly affect continuance intention via attitude. These findings made the grounds that perceived usefulness and perceived enjoyment need to be chosen as the leading factors in understanding the continuance intention of AIPA. TAM was developed to explain the acceptance of technology []. According to the model, the easier and more useful the technology is, the stronger the user's intention to accept it. Technology acceptance factors have been also verified to enhance continuance intention []. Although the research model in has an advantage in that it integrates traditional models in the field of ISs, it did not present AIPA-specific variables. shed light on the shopping intention at AI-controlled retail stores. They showed that innovativeness and optimism of consumers are the major antecedents of perceived ease and perceived usefulness. Also, it was revealed that perceived ease of use, perceived usefulness, perceived enjoyment, customization, and interactivity are significant predictors of the shopping intention of consumers in AI-powered automated stores. Based on the results of , this study reflects perceived ease of use, perceived usefulness, and perceived enjoyment. Since dealt with AI-powered stores. the research results may be limitedly applicable to the shopping context. To get a broader perspective, this paper aimed to study general AIPA, which is most easily encountered by people. It will derive implications that can be applied to detailed AI subjects [e.g., shopping, game, education, etc.]. investigated the decision factors of usage in the domain of virtual personal assistants. They pointed out that parasocial interaction, assistant type, and loneliness play a crucial role in enhancing the level of usage. On the ground of above results [; ], this study introduces parasocial interaction as the key factor in the formation of continuance intention. has limitations in that it does not consider the function, technology, and value of AIPA. To overcome it, this study measured related factors from the users' cognitive perspective.
identified the determinants of brand loyalty in the context of voice-assisted AI. They found trust, interaction, and novelty value are the significant and positive precursors of brand loyalty. Perceived risk was found to hurt brand loyalty. Brand involvement and consumer innovativeness moderated the effect of novelty value on brand loyalty positively. Based on the results of , this study employs novelty value in elaborating continuance intention. Since explained brand loyalty by examining only Siri, the result would be applicable to only a certain brand among many AIPAs. To improve the generality, this study describes the general intention of using AIPA by investigating multiple assistants. Also, suggested basic interaction as the antecedent of brand loyalty. This article introduces parasocial interaction based on the human-like behavior of AIPA. investigated the effect of privacy concerns on the use of a voice-activated personal assistant in the public space. They observed the behavior of users according to location and the type of information. The authors found that users are more likely to transmit information in private spaces than in public spaces. Users also were figured out to be more likely to transmit non-private information than private information. examined the roles of task complexity and problem-solving ability in shaping usage intention in AI customer service. They intimated that the consumers considered the problem-solving ability of AI to be greater than that of human customer service in low-complexity tasks. Perceived problem-solving ability mediated the effects of customers’ service usage intentions with task complexity serving as a boundary condition. included only task-complexity and problem-solving ability because the subjects of the study were used only for utility purposes. Since AIPA can provide both utilitarian value and hedonic value, this study considered both utility and hedonic aspects. Table 1 describes a critical overview of related studies.
Table 1
A critical overview of related works.
StudyMethodologyContext/SettingSample frameSample SizeMain VariablesOutcomesCritical OverviewCross-sectional; surveySmart voice assistantAlexa users and Google Assistant users244User engagement; trust; privacy risk; satisfaction; slowness of adoption; skepticism; attitudeContinuance usage reflected the utilitarian attitude and the hedonic attitude as exogenous variables. On the other hand, the current study observes the formation process of continuance intention more elaborately by presenting the evidence factors that determine the utilitarian value and the hedonic value.Cross-sectional; surveyAI In-home voice assistantAmazon Echo users724Utilitarian benefits; hedonic benefits; symbolic benefits; social presence; social interaction; perceived privacy riskUsage explicated the use of voice assistants based on utilitarian benefits and hedonic benefits. The present research approaches the user behavior of AIPA more delicately by observing the formation mechanism of utilitarian value and hedonic value.Cross-sectional; surveyChatbotBank's chatbot users359Information quality; system quality; service quality; trust; user satisfaction; confirmation expectations; perceived usefulnessContinuance intention reflected robust variables, but they have been overused in IT contexts. The current study is different in that it includes AIPA-specific variables while maintaining the major determinants of IT use.Cross-sectional; surveyVoice-user interfaceVoice-user interface users414Gender; information quality; information satisfaction; system quality; system satisfaction; perceived usefulness; perceived ease of use; perceived enjoyment; mobile self-efficacy; trust; perceived risk; attitudeContinuance intention introduced representative variables of the IS success model and technology acceptance model. They did not reflect the unique characteristics of AIPA, which communicates through voice. The present research employs AIPA-specific factors and systematically structured them into two aspects: utilitarian value and hedonic value.Cross-sectional; surveyAI-powered automated retail storesAI-powered store consumers1250Optimism; innovativeness; discomfort; insecurity; perceived usefulness; perceived ease of use; perceived enjoyment; customization; interactivityIntention to shop focused on AI in the shopping environment by considering the tendencies of consumers. On the other hand, the current work aims to study AIPA, which is most easily encountered by people. It can derive implications that can be applied to detailed AI subjects [e.g., shopping, game, education, etc.].Cross-sectional; surveyVirtual personal assistantSmart speaker users534Parasocial interaction; Personification type; LonelinessSatisfaction investigated only how parasocial interaction, types of assistants, and loneliness affect satisfaction. The present paper has limitations in that it does not consider the function, technology, and value of assistants. To overcome it, this study measured related factors from the users' cognitive perspective.Cross-sectional; surveyVoice-controlled AISiri users675Trust; interaction; perceived risk; novelty value; employment; brand involvement; consumer innovativenessBrand loyaltyWhile explained brand loyalty by examining only Siri, the current article describes the general intention of using AIPA by investigating multiple assistants. suggested basic interaction as the antecedent of brand loyalty. This study introducesparasocial interaction based on the human-like behavior of AIPA.Study 3; Experimental studyAI customer serviceBank's AI Online service51Online customer service; perceived problem-solving ability; task-complexityUsage intentionIn explaining the use of AI customer service, considered only task complexity and problem-solving ability. The subjects of the study are used only for utility purposes. Because AIPA can provide both utilitarian value and hedonic value, this study considered both utility and hedonic aspects.
In summary, a large body of studies has applied the IS Success Model, TAM, and ECM, which have been widely verified in the IS field. Moreover, some researchers have introduced variables such as perceived risk, trust, interaction, and problem-solving to explain the behavior of AIPA users. Based on previous works, the current study employs the factors that precede the intention to continue using AIPA.
3. Research model
Figure 1 depicts the research model to identify the predictors of continuance intention in the context of AIPA. AIPA provides useful functions and answers interesting command results. Thus, this study suggests that utilitarian value and hedonic value significantly affect continuance intention. Utilitarian value may increase if users can use helpful features from AIPAs and handle them comfortably. Hence, this research postulates that perceived ease of use and perceived usefulness are significant predictors of utilitarian value. Hedonic value may be enhanced if users are entertained with AIPA and interact with it similarly to humans. Therefore, this article posits that perceived enjoyment and parasocial interaction significantly impact hedonic value. AIPA's novelty value includes usefulness and fun. Consequently, novelty value is expected to affect both utilitarian value and hedonic value.
3.1. Utilitarian value [UTV]
Utilitarian value reflects the task-specific, efficient, and economical aspects of a product/service []. Thus, the continuous use of AIPA can be regarded as a means of accomplishing some task-related goals []. Utilitarian benefits are related to useful help. Utilitarian value has been figured out to be the determinant of behavioral intention in the various IS contexts [; ; ]. Users perform useful functions such as registering schedules or searching for something through AIPA. Users with a higher level of utilitarian value would increase their intention to continue using AIPA. Thus, this study predicts that utilitarian values facilitate the intention to use continuously.
H1. Utilitarian value significantly affects continuance intention.
3.2. Hedonic value [HEV]
Hedonic value represents enjoyment, pleasure, and anxiety related to the use of a product/service []. It can be drawn from the interaction with technology itself []. Hedonic value significantly determines the adoption intention or continuance intention in the IS domains [; ; ; ]. AIPA provides fun through voice interaction with humans. Users that gain greater hedonic value are more motivated to use AIPA. Therefore, one can expect that hedonic value triggers continuance intention.
H2. Hedonic value significantly affects continuance intention.
3.3. Perceived ease of use [PEU]
Perceived ease of use refers to the extent to which a person believes that using a particular system would be free of effort []. It significantly drives the utilitarian performance expectancy of mobile IT []. Perceived ease of use is the leading factor of utilitarian value []. AIPA helps the user by processing the voice requested by the user. Perceived ease of use in the AIPA context may be related to the recognition of pronunciation and command understanding []. Users may get useful help if AIPA recognizes spoken words more accurately and performs tasks better. Easy AIPA will increase the utilitarian value of users. As a consequence, perceived ease of use is believed to positively control utilitarian value.
H3. Perceived ease of use significantly affects utilitarian value.
3.4. Perceived usefulness [PUS]
Perceived usefulness is described as the user's belief about whether their experiences are improved by using technology []. It accelerates attitude or continuance intention in the domain of AIPA [; , ]. If AI offers users higher levels of perceived usefulness, their shopping intention increases in AI-powered stores []. AIPA may play an important role in purchasing decisions by mediating relationships between brands, retailers, and consumers []. The more useful help users get from AIPA, the higher they will assess. Hence, this study proposes that perceived usefulness elevates the level of utilitarian value.
H4. Perceived usefulness significantly affects utilitarian value.
3.5. Novelty value [NOV]
Novelty value refers to the degree to which a product/service differs from others in terms of originality and uniqueness []. It significantly forms brand loyalty toward voice-controlled AI []. Unlike existing PCs and smart devices, AIPA provides useful help and fun through voice recognition. Users with a higher degree of novelty value may increase the level of utilitarian value and hedonic value. Based on this, novelty value is suggested to enhance the levels of utilitarian value and hedonic value.
H5a. Novelty value significantly affects utilitarian value.
H5b. Novelty value significantly affects hedonic value.
3.6. Perceived enjoyment [PEN]
Perceived enjoyment is defined as the extent to which the activity of using a specific IT is perceived to be enjoyable in its own right []. It is significantly associated with the continuance intention of hedonic IS []. When users perceive AIPA as more enjoyable, their attitudes toward AIPA are increased []. Users may find pleasure in having a conversation with a device rather than a person. Perceived social interaction with AIPA is measured by the pleasure of conversation []. This fun may provide users with hedonic value. Therefore, one can expect that perceived enjoyment leverage hedonic value.
H6. Perceived enjoyment significantly affects hedonic value.
3.7. Parasocial interaction [PSI]
Parasocial interaction is related to personal relationships, responsiveness, reality, and friendliness [; ]. Parasocial relationship enhances users' satisfaction and continuance intention of users in the context of intelligent personal assistant []. It has a significant correlation with perceived enjoyment []. As AIPA provides customized answers to users and performs interactive communication, parasocial interaction is formed between the user and the assistant. This sort of human-like communication would deliver hedonic value. In this vein, parasocial interaction is expected to raise the level of hedonic value.
H7. Parasocial Interaction significantly affects hedonic value.
4. Methodology
4.1. Measurement instrument
All measurement indicators were drawn from previously verified studies in IS and AI fields. The survey questions were modified to suit the AIPA context. All items were measured based on a 7-point Likert scale ranging from 1 [strongly disagree] to 7 [strongly agree]. A 7-point Likert scale is more likely to generate slightly higher mean scores as compared to a 10-point scale, which makes comparing data a much easier process []. Moreover, it has been shown to provide higher sensitivity and better discrimination between participants []. Before survey implementation, three researchers in the IS and quantitative analysis reviewed it, assuring content validity, logical arrangement, and question ambiguity. To measure each dimension effectively in this study, the operational definitions are formulated and shown below in Table 2. Also, the measurement items and sources are presented in Table A1.
Table 2
Operational definition.
ConstructOperational DefinitionNumber of QuestionsReferenceContinuance IntentionUser's overall willingness to continue using AIPA3Utilitarian ValueThe degree to which AIPA provides beneficial and good value3Hedonic ValueThe degree to which AIPA provides pleasure, motivation, and comfort3Perceived Ease of UseThe extent to which AIPA is clear, understandable, and easy to use3Perceived UsefulnessThe extent to which the use of AIPA is useful and helps users complete work3Novelty ValueThe degree to which AIPA delivers new experiences and satisfies user curiosity3Perceived EnjoymentThe extent to which AIPA is enjoyable and interesting to use3Parasocial InteractionThe degree to which users perceive communication with AIPA realistically and personally3
4.2. Subject and data collection
The research model was empirically tested by the use of data gathered from a cross-sectional survey. Ethical approval for the survey was obtained from the Ethical Committee of RealSecu. Data were collected from a commercial online survey provider with a large number of panels in South Korea because of low sample bias. The online link to access the questionnaire was distributed to its panels. Respondents were rewarded USD 4.18. To validate users, the questionnaire first asked whether they had used an AIPA. If their answer was affirmative, they were allowed to fill out questions. Only after answering all questions on each page could participants move to the next page. After discarding insincere responses, 257 data were utilized for the next analysis. This study checked the adequate sample size for structural equation modeling. A priori sample size calculator was used to confirm the minimum requirement for structural equation modeling []. Inputting the required information such as 0.1 anticipated effect size, 80% desired statistical power level, 8 number of latent variables, 24 number of observed variables as well as 0.05 probability level, the minimum required sample size is 200. Since the sample size of this study is 257, this requirement is met as well. Among the final samples, 52.1% were male and 47.9% were female. The mean age of the final sample was 35.1 with a standard deviation of 9.44. 102 respondents were using Siri and 124 respondents were using Bixby. The distribution of the subjects of the current study is different compared to the global market share []. This is probably because Koreans prefer Samsung which is a Korean company and manufactures Bixby. 102 informants were using iPhones and 149 informants were using Samsung phones. Table 3 shows the demographic information of respondents in the final sample.
Table 3
Profile of respondents.
DemographicsItemSubjects [N = 257]
FrequencyPercentageGenderMale13452.1%Female12347.9%Age10s103.9%20s6926.8%30s9336.2%40s6424.9%50s217.3%AIPA TypeSiri10235.7%Bixby12443.4%Alexa20.7%Google Assistant279.4%Etc20.7%Number of uses