Original Manuscript

Information Development
1–21
Decoding the ChatGPT mystery: A © The Author(s) 2023
Article reuse guidelines:

comprehensive exploration of factors sagepub.com/journals-permissions

DOI: 10.1177/02666669231202764
journals.sagepub.com/home/idv
driving AI language model adoption

Hyeon Jo
HJ Institute of Technology and Management, Bucheon, Republic of Korea

Abstract
The increasing ubiquity of Artificial Intelligence (AI) chatbots across a variety of sectors has sparked a burgeon-
ing interest in deciphering the determinants that govern their adoption and usage. This study aims to examine
the pivotal factors that influence the practical usage of the AI chatbot, ChatGPT, among a sample of university
students. Leveraging the theoretical framework of planned behavior, the research model scrutinizes the inter-
play between knowledge application, perceived intelligence, usability, attitude, subjective norms, perceived
behavioral control, trust, behavioral intention, and actual usage. Data procured from a survey of university
students were examined through the lens of structural equation modeling. The outcomes reveal that knowl-
edge application, perceived intelligence, and usability have a positive impact on attitudes towards ChatGPT.
Perceived intelligence also influences knowledge application, usability, and trust. Concurrently, attitude and
subjective norms notably affect behavioral intention. Contrary to expectations, perceived behavioral control
did not significantly influence behavioral intention. Trust emerged as a crucial factor steering behavioral inten-
tion, which in turn, positively correlates with the actual usage of ChatGPT. These insights enrich the academic
discourse on AI chatbot adoption and provide practical implications for AI developers, educators, and policy
makers, striving to enhance user engagement with AI systems in educational contexts.

Keywords
ChatGPT, AI language model, knowledge application, perceived intelligence, trust

Submitted: 25 May 2023; accepted: 5 September 2023.

Introduction engaging in natural, human-like conversation. While

AI language models like ChatGPT can underpin chat-
Rapid advancements in artificial intelligence (AI) have
bots, not all chatbots incorporate such models. Some
spurred the development of increasingly sophisticated
resort to simpler, rule-based systems for response gen-
and adaptable chatbots, finding utility in various indus-
eration. However, advanced chatbots, such as those
tries, including education (Al-Emran et al., 2023; Tlili
built on ChatGPT, utilize AI language models to gener-
et al., 2023). A chatbot is software engineered for auto-
ate more complex, contextually appropriate responses,
mating human interactions (Lebeuf et al., 2017). One
thereby enhancing the user experience (Brandtzaeg &
such AI chatbot, garnering significant attention from
Følstad, 2017). According to a study, the majority of
researchers, educators, and students alike, is
students (89%) reported using ChatGPT to assist
ChatGPT, developed by OpenAI (Farrokhnia et al.,
2023; Lund et al., 2023). This bot is also an AI language
model, predictive in nature, and generates sequences of
Corresponding author:
words using machine learning based on the provided
Hyeon Jo, Headquarters, HJ Institute of Technology and
input (Devlin et al., 2018; Radford et al., 2019). It Management, 71 Jungdong-ro 39, Bucheon-si, Gyeonggi-do,
refines its responses through pattern recognition in 14721, Republic of Korea.
data, making it an advanced AI language model for Email: [email protected]
2 Information Development 0(0)

them with their homework assignments (Study.com, new dimension to the existing body of research, sig-
2023). Given its potential to transform the way students nificantly advancing our understanding of AI chatbot
engage with learning materials and interact with their adoption in educational settings.
peers and educators, understanding the factors that Research on technology adoption in education pri-
drive its adoption and usage among university students marily uses the technology acceptance model (TAM)
has become crucial. This study aims to investigate the and theory of planned behavior (TPB) to understand
determinants of ChatGPT adoption among university AI chatbot user behavior (Chang et al., 2022). TAM
students and explore the underlying mechanisms that suggests users’ attitudes towards technology are
influence their actual usage of this AI chatbot. shaped by perceived usefulness and ease of use. In
The adoption of AI chatbots, such as ChatGPT, in the context of education, perceived usefulness means
the educational sector has been met with enthusiasm believing a technology, like ChatGPT, will enhance
and skepticism (Dwivedi et al., 2023; Lo, 2023). On learning outcomes. This can extend to knowledge
the one hand, proponents argue that AI chatbots can application, reflecting how students feel using
revolutionize the learning experience by offering per- ChatGPT benefits their learning and problem-solving.
sonalized learning paths, instant feedback, and A critical dimension is perceived intelligence, captur-
round-the-clock assistance, thereby enhancing stu- ing users’ belief in ChatGPT’s intelligence level in
dents’ motivation, engagement, and academic per- aiding their tasks. Higher perceived intelligence
formance (Sullivan et al., 2023). On the other hand, often leads to favorable attitudes towards AI chatbots
skeptics express concerns regarding the potential (Rafiq et al., 2022). However, ChatGPT has faced
loss of human touch, privacy issues, and the risk of technical issues affecting usability, such as chat
reinforcing biases and misinformation (Cotton et al., history misses (Joseph, 2023). As usability is vital in
2023; Johnson et al., 2023). Thus, it becomes essential educational information systems (Söderlund, 2022),
to understand the factors that contribute to the adop- it’s essential for designing chatbots meeting varied
tion of ChatGPT by university students to maximize user requirements and promoting positive usage atti-
its benefits while mitigating potential drawbacks. tudes. Trust in information security is pivotal for tech-
Despite the growing interest in ChatGPT, empirical nology adoption. Users trust systems they believe
studies exploring the factors influencing the adoption protect their data (Ho et al., 2017). For ChatGPT,
of a large language model in education remain sparse. this means users’ confidence in its ability to prevent
While several studies have investigated chatbot unauthorized data access. A glitch revealing
acceptance in education (Hew et al., 2021; Mohd ChatGPT users’ information (Zorz, 2023) highlighted
Rahim et al., 2022; Sandu & Gide, 2019) and other trust importance. If compromised, students might
sectors such as customer service and e-commerce avoid or abandon the platform, impacting their learn-
(Chung et al., 2020; Følstad et al., 2018; Pillai and ing and ChatGPT’s educational adoption. Trust in
Sivathanu, 2020), the distinctive characteristics of a security, thus, becomes a key factor in user intentions
large language model and the unique demands of edu- toward ChatGPT, as it minimizes perceived risks and
cational environments necessitate targeted research. encourages AI chatbot interactions.
Most existing studies primarily focus on technology Although there is an increasing body of research on
acceptance models, investigating factors like per- the factors influencing the adoption and usage of AI
ceived usefulness and ease of use in relation to chatbots in various contexts, several research gaps still
chatbot adoption (Ashfaq et al., 2020; Nguyen et al., exist, particularly in the context of university students’
2021). However, there is a paucity of studies that use of ChatGPT. While prior studies have explored AI
delve into AI language model-specific factors, such chatbot adoption in diverse domains such as healthcare,
as perceived intelligence, trust in information security, customer service, and e-commerce (Chung et al., 2020;
and knowledge application. Moreover, studies explor- Mokhtar and Salimon, 2022; Zhang et al., 2023), there is
ing the impact of social influences, such as subjective a lack of research specifically targeting university stu-
norms, on chatbot adoption in educational settings are dents. Moreover, the role of knowledge application, per-
notably scarce. This study seeks to address these gaps ceived intelligence, usability, and trust, essential aspects
by providing empirical evidence on the determinants of user experience, has received limited attention. By
of AI chatbot adoption among university students. addressing these research gaps, the present study seeks
By examining factors such as perceived intelligence to contribute to the AI chatbot adoption literature by pro-
and knowledge application, the study introduces a viding a more nuanced understanding of the factors
Jo: Decoding the ChatGPT mystery 3

influencing university students’ attitudes and intentions Knowledge acquisition

to use ChatGPT. Therefore, the research questions of
The adoption of knowledge application as a key deter-
this paper are:
minant of attitude toward ChatGPT can be justified by
drawing upon several theoretical frameworks and
1. Can TPB be applied to explain the behaviors of
empirical evidence in the technology adoption litera-
students using ChatGPT?
ture. In particular, the TAM and the TPB provide
2. Does knowledge application, perceived intelli-
strong theoretical foundations for the inclusion of
gence, and usability impact attitudes toward
knowledge application in the research model.
ChatGPT, and does trust influence the behav-
The TAM posits that users’ attitudes toward technol-
ioral intention to use ChatGPT?
ogy are influenced by two primary factors: perceived
usefulness and perceived ease of use. Perceived useful-
Theoretical background ness refers to the degree to which users believe that
TPB adopting a particular technology will enhance their job
performance or, in the context of education, their learn-
The TPB is a widely used psychological framework for
ing outcomes. In this regard, knowledge application can
understanding human behavior and decision-making
be considered as an extension of perceived usefulness,
processes (Ajzen, 1991). It posits that an individual’s
as it specifically captures the extent to which students
behavior is determined by their behavioral intentions,
believe that using ChatGPT will improve their learning
which are in turn influenced by attitudes, subjective
and problem-solving abilities. The TPB also emphasizes
norms, and perceived behavioral control (PBC). TPB
the role of beliefs in shaping attitudes and intentions.
has been successfully applied in various contexts (Ang
According to the TPB, individuals’ attitudes are
et al., 2021; Huang, 2023; Song and Jo, 2023), including
formed based on their beliefs about the consequences
technology adoption and usage, making it a suitable the-
of engaging in a specific behavior, in this case, using
oretical foundation for investigating students’ behaviors
ChatGPT. By incorporating knowledge application
in using ChatGPT.
into the research model, we can better capture students’
In the context of ChatGPT, students’ attitudes towards
beliefs about the potential benefits of using ChatGPT for
using the AI chatbot are shaped by their beliefs about the
enhancing their learning experience.
benefits and drawbacks of using the technology. The
Empirical evidence further supports the significant
TPB suggests that when students have favorable attitudes
influence of knowledge application on users’ attitudes
towards ChatGPT, they are more likely to form strong
and intentions to adopt educational technologies
intentions to use it. Subjective norms may play a role in
(Al-Emran and Mezhuyev, 2020; Al-Emran and Teo,
determining students’ intentions to use the chatbot
2020). By incorporating knowledge application into
based on the influence of their social networks. If students
the research model, this study acknowledges the
perceive that using ChatGPT is socially desirable and
unique features of AI chatbots, such as ChatGPT,
encouraged by others, they are more likely to form posi-
that distinguish them from other educational technolo-
tive behavioral intentions. Lastly, PBC may involve stu-
gies. Specifically, ChatGPT’s ability to engage in
dents’ self-confidence in navigating and interacting with
natural language conversation, provide instant feed-
the AI chatbot effectively. When students perceive that
back, and facilitate personalized learning experiences
they have the necessary skills and resources to use
makes the adoption of knowledge application particu-
ChatGPT, they are more likely to form stronger inten-
larly relevant for understanding students’ attitudes
tions to adopt the technology.
toward ChatGPT.
By adopting the TPB as the theoretical basis, this study
To sum up, the decision to incorporate knowledge
can provide insights into the factors influencing univer-
application as a factor in explaining attitude toward
sity students’ intentions and actual usage of ChatGPT.
ChatGPT is firmly rooted in the TAM and TPB frame-
Moreover, it allows for the exploration of the interplay
works, as well as empirical findings from previous studies.
among attitudes, subjective norms, and PBC, as well as
their individual effects on students’ adoption of AI chat-
bots in an educational setting. Understanding these rela- Perceived intelligence
tionships can contribute to the development of strategies The TPB highlights the role of beliefs in shaping atti-
for enhancing ChatGPT adoption and improving its tudes and intentions. According to the TPB, indivi-
effectiveness as a learning tool. duals’ attitudes are formed based on their beliefs
4 Information Development 0(0)

about the consequences of engaging in a specific of acceptance and use of technology (UTAUT), intro-
behavior, in this case, using ChatGPT. By incorporat- duced by Venkatesh et al. (2003), is an extension of
ing perceived intelligence into the research model, we the TAM that incorporates additional determinants
can better capture students’ beliefs about the potential of technology acceptance. One of the key constructs
benefits of using an intelligent chatbot like ChatGPT in UTAUT is facilitating conditions, which refer to
for enhancing their learning experience. the degree to which users believe that organizational
Considering the TAM, perceived intelligence can be and technical infrastructure supports the use of a
viewed as an extension of perceived usefulness, as it given technology. By ensuring that ChatGPT is
captures the degree to which users believe that more usable, it is more likely to be perceived as valu-
ChatGPT’s intelligence level contributes to its effective- able and efficient, which, in turn, can positively influ-
ness in supporting their learning and problem-solving ence attitudes toward its adoption. Empirical evidence
tasks. According to the TPB, individuals’ attitudes are also supports the inclusion of usability in the research
formed based on their beliefs about the consequences model. Numerous studies have demonstrated the sig-
of engaging in a specific behavior, in this case, using nificant impact of usability on users’ attitudes, satis-
ChatGPT. faction, and technology adoption intentions (Dubey
The computers are social actors (CASA) paradigm, and Sahu, 2021; Tsai et al., 2014; Yan et al., 2023).
developed by Reeves and Nass (1996), posits that To conclude, the rationale for incorporating
humans tend to interact with computers, and by exten- usability as a factor in explaining attitude toward
sion, AI chatbots, as if they were social actors. This ChatGPT is firmly rooted in the TAM and
implies that users’ perceptions of ChatGPT’s intelli- UTAUT, along with empirical findings from previ-
gence can significantly influence their attitudes ous studies.
toward the chatbot, as they may attribute human-like
qualities to it. In this context, higher perceived intelli-
gence of ChatGPT may lead to more positive attitudes Trust
as users might believe that a more intelligent chatbot Trust in information security is an essential factor in
can better understand their needs, provide more accur- technology adoption, as it determines users’ willing-
ate information, and offer more effective support in ness to engage with a system based on their confidence
their learning process. in the protection of their data (McKnight, 2005). In the
Empirical evidence further supports the inclusion of context of ChatGPT, trust in information security is
perceived intelligence in the research model. Several particularly relevant because users may share sensitive
studies have demonstrated the significant influence of and personal information through their interactions
system intelligence on users’ attitudes and intentions with the AI language model, especially in educational
to adopt various technologies (Rafiq et al., 2022; settings. The privacy calculus theory posits that indivi-
Tussyadiah and Park, 2018). In the context of AI chat- duals assess the potential benefits and risks associated
bots, research has shown that perceived intelligence with disclosing their information when using technol-
plays a crucial role in shaping users’ satisfaction, trust, ogy (Dinev and Hart, 2006; Laufer and Wolfe, 1977).
and adoption intentions (Alaiad and Zhou, 2014; When users perceive that their information is secure
Huang and Rust, 2018). and the benefits of using the technology outweigh the
In summary, the rationale for incorporating per- risks, they are more likely to adopt the technology.
ceived intelligence as a determinant of attitude Furthermore, trust has been identified as a critical deter-
toward ChatGPT is firmly anchored in the TAM, minant of behavioral intention in the modified or
TPB, and CASA theories, as well as empirical find- extended models originating from the expectation-
ings from earlier investigations. confirmation model (ECM) (Bhattacherjee, 2001;
Brill et al., 2019), TAM (Gefen et al., 2003), and the
UTAUT (Patil et al., 2020; Slade et al., 2015). As
Usability well, literature on AI has stressed trust in explicating
Following the TAM, usability can be closely related to satisfaction and behavioral intention (Brill et al.,
the perceived ease of use, as it refers to the extent to 2019; Nguyen et al., 2019, 2021; Pal et al., 2021).
which users can effortlessly interact with ChatGPT, These models emphasize the importance of trust in fos-
regardless of their individual abilities, technical tering user engagement and facilitating the adoption
skills, or situational constraints. The unified theory process.
Jo: Decoding the ChatGPT mystery 5

In light of these theoretical underpinnings, it is 2020). According to the TAM (Davis, 1989), perceived
crucial to incorporate trust in information security as a usefulness, which includes knowledge application, is a
determinant of behavioral intention toward ChatGPT. key determinant of an individual’s attitude toward tech-
nology adoption. In line with the TAM, previous research
Conceptual model and research has shown that when students perceive technology as
hypotheses useful for applying their knowledge, they are more
likely to develop a positive attitude toward its adoption
The research model for this study integrates key con-
(De Cicco et al., 2022; Nadarzynski et al., 2019).
structs from well-established theories to investigate
Consequently, this study hypothesizes the following:
the factors influencing university students’ attitudes
and intentions to adopt ChatGPT in their learning H1: Knowledge application has a positive effect on
process. The model comprises the following con- attitude.
structs: knowledge application, perceived intelligence,
usability, attitude, subjective norms, PBC, trust in
Perceived intelligence
information security, behavioral intention, and actual Perceived intelligence refers to an individual’s appraisal
usage. of a technology’s capability to exhibit intelligent beha-
Knowledge application, perceived intelligence, and viors, such as delivering reasonable replies, transmitting
usability are proposed as antecedents of attitude edge data, and understanding complex scenarios
toward ChatGPT. Attitude, subjective norms, and (McGinn et al., 2019; Petisca et al., 2015). Research sug-
PBC are hypothesized to influence behavioral inten- gests that the perceived intelligence of AI-based applica-
tion, while trust in information security is expected tions determines their practical usefulness (Moussawi and
to have a direct effect on behavioral intention. Koufaris, 2019; Moussawi et al., 2021). Intelligent
Lastly, behavioral intention is posited to have a posi- systems are expected to comprehend and process infor-
tive impact on the actual usage of ChatGPT among mation more effectively, facilitating knowledge applica-
university students Figure 1. tion (Sarker, 2022). Specifically, in academia, systems
like ChatGPT, when perceived as intelligent, can
enhance learning processes and promote knowledge
Knowledge application application among students. The perception of an AI
Knowledge application refers to the ability to utilize system’s intelligence is closely related to its perceived
acquired knowledge effectively in various tasks, such as usability (Moussawi et al., 2021). Intelligent systems
problem-solving, decision-making, and collaboration like ChatGPT can potentially reduce complexity and
(Alavi and Leidner, 2001). The importance of knowledge enhance interactions, improving overall usability
application in technology adoption, particularly in an edu- (Sundar et al., 2014). Past research emphasizes perceived
cational context, has been emphasized in recent studies intelligence’s role in shaping user attitudes towards tech-
(Al-Emran and Mezhuyev, 2020; Al-Emran and Teo, nology adoption (Balakrishnan and Dwivedi, 2021;

Figure 1. Research framework.

6 Information Development 0(0)

Tussyadiah and Park, 2018). Studies show that users who (Al-Emran et al., 2018; Venkatesh and Davis, 2000).
perceive a technology as intelligent are more likely to When users have a positive attitude toward technology,
develop positive attitudes towards its use (Rafiq et al., they are more likely to develop an intention to use it.
2022), directly impacting their adoption decisions. This is because a favorable attitude enhances the per-
When users perceive an AI-driven chatbot, like ceived benefits of adopting the technology, thereby
ChatGPT, as highly intelligent, they may believe that increasing the likelihood of adoption (Dubey and Sahu,
the system employs advanced mechanisms ensuring 2021; Kai–ming Au and Enderwick, 2000; Sciarelli
information security (Alaiad and Zhou, 2014). The per- et al., 2022). A favorable attitude toward AI artifacts facil-
ceived intelligence may lead users to believe in the AI itates behavioral intention such as continuance intention
system’s ability to effectively detect and prevent security (Nguyen et al., 2019; Pal et al., 2021) or recommendation
breaches, thereby enhancing their trust in its security mea- intention (Mishra et al., 2022). Accordingly, this paper
sures (Huang and Rust, 2018). Users perceiving suggests the following:
ChatGPT as highly intelligent might be more confident
in its ability to manage their information securely, H4: Attitude has a positive effect on behavioral
thereby fostering increased trust. Hence, this paper intention.
hypothesizes:
Subjective norms
H2a: Perceived intelligence has a positive effect on
knowledge application Subjective norms refer to an individual’s perception of
H2b: Perceived intelligence has a positive effect on social pressure to perform or not perform a specific
usability. behavior (Ajzen, 1991). Recent empirical studies
H2c: Perceived intelligence has a positive effect on have demonstrated the significant influence of subject-
attitude. ive norms on behavioral intention in various contexts,
H2d: Perceived intelligence has a positive effect on including technology adoption (Bruce et al., 2023;
trust. Buba et al., 2022; Sun et al., 2020). Users are more
likely to develop a behavioral intention to use a tech-
Usability nology if they perceive that important referents, such
Usability, in the context of this study, refers to the ease as friends, family, or colleagues, approve of its
with which users can access, interact with, and benefit usage (Rachmawati et al., 2020). The higher level of
from the system (Almaiah and Man, 2016; Borsci subjective norms, the more users are likely to adopt
et al., 2022; Rafiq et al., 2022). When users find a AI chatbots (Belanche et al., 2019; Gatzioufa and
chatbot easily accessible, they are more likely to Saprikis, 2022). Thus, positive subjective norms
develop positive attitudes toward it (Yoon and Yu, regarding ChatGPT can be expected to contribute to
2022). Research demonstrates that perceived usability a higher intention to use the system among users.
significantly enhances both cognitive and affective atti-
tudes, as users typically favor systems that are easy to H5: Subjective norms have a positive effect on behav-
use and navigate (Rafiq et al., 2022). Improved usability ioral intention.
can contribute to increased satisfaction, which in turn,
leads to a more favorable attitude toward the technology. Perceived behavioral control
If users perceive the ChatGPT as easily accessible, they Perceived behavioral control refers to an individual’s
are more likely to appreciate its features and functional- belief in their ability to execute the behavior required
ities, resulting in a positive attitude towards its adoption. to use a system based on their perceived ease or diffi-
Therefore, this study proposes the following: culty in performing the behavior (Ajzen, 1991).
Recent research supports the positive relationship
H3: Usability has a positive effect on attitude. between perceived behavioral control and behavioral
intention in technology adoption contexts (Chao,
Attitude 2019; Hansen et al., 2018). Higher levels of perceived
Attitude refers to an individual’s overall evaluation of behavioral control can lead to a stronger intention to
performing a specific behavior, such as using ChatGPT use technology as users perceive that they have the
(Ajzen, 1991). Numerous studies have supported the necessary resources and opportunities to engage with
link between a positive attitude and increased behavioral AI systems (Chai et al., 2020). Consequently, this
intention in the context of technology adoption study posits that greater perceived behavioral control
Jo: Decoding the ChatGPT mystery 7

regarding ChatGPT will result in a higher intention to 2003). Men and women may perceive and interact
use the technology among users. with technology differently, resulting in variations in
their attitudes and intentions toward adopting a spe-
H6: Perceived behavioral control has a positive effect on cific technology like ChatGPT. For example, women
behavioral intention. are more influenced by subjective norms and social
Trust factors, whereas men tend to be more driven by per-
ceived usefulness and ease of use (Venkatesh et al.,
Trust is regarded as a critical factor in adopting tech- 2003). Age, on the other hand, can affect technology
nology (Ho et al., 2017; Lin et al., 2020). Recent adoption through differences in cognitive abilities,
studies have demonstrated a positive relationship prior experience, and technology-related attitudes
between trust and behavioral intention in various tech- (Morris and Venkatesh, 2000). Older individuals
nology adoption contexts (AlHogail, 2018; Hooda may have different expectations, preferences, and
et al., 2022; Wang et al., 2015). Trust can alleviate technological skills compared to younger users,
uncertainty and perceived risks, leading to higher sat- which can shape their perceptions and attitudes
isfaction with AI applications (Brill et al., 2019). toward adopting ChatGPT. Moreover, younger users
Perceived risk in information security harms the atti- are often considered digital natives and may have
tude toward AI assistants (Nguyen et al., 2019), and more familiarity with emerging technologies, poten-
indirectly influences the continuance intention of AI tially affecting their adoption behaviors.
assistants (Han and Yang, 2018). In light of the exist- By including gender and age as control variables,
ing literature, it is reasonable to suggest that a higher the study aims to account for the potential effects of
level of trust in ChatGPT will result in a greater inten- these individual differences on the relationships
tion to use the technology among users. Therefore, between the main constructs of the research model.
this study puts forth the hypothesis that trusts posi-
tively impacts behavioral intention.

H7: Trust has a positive effect on behavioral intention. Empirical methodology

Measurement instrument
To ensure the validity and reliability of the measure-
Behavioral intention ment scales used in this study, we derived the instru-
ments from prior validated research. Alongside these
Behavioral intention refers to the likelihood that an indi-
scales, we included demographic questions in our
vidual will perform a specific behavior or use a particular
survey to gather information about participants’ age
technology (Ajzen, 1991). Numerous studies in the field
and gender. The questionnaire was initially prepared
of technology adoption have reported a positive relation-
in English by the authors. The English version was
ship between behavioral intention and actual usage
translated into Korean by a bilingual researcher in
across diverse contexts (Khan and Qudrat-Ullah, 2021;
the information technology area. Then, the Korean
Agudo-Peregrina et al., 2014; Lai et al., 2022). As stu-
version was back-translated into English by a bilin-
dents’ intentions to continuously use the educational
gual professional. The two English versions of the
information system increase, their actual usage increases
questionnaire had only slight differences that were
(Al-Emran et al., 2020). Given the existing evidence, it is
adjusted by the author. The initial pool of items was
reasonable to expect that a stronger intention to use
reviewed by a panel of experts consisting of research-
ChatGPT among users will lead to a higher likelihood
ers and practitioners familiar with the topic of technol-
of actual usage. Consequently, this study suggests that
ogy adoption and the context of AI chatbots in
behavioral intention positively affects the actual usage
education. The panel assessed the content validity of
of ChatGPT.
the items by examining their clarity, relevance, and
H8: Behavioral intention has a positive effect on actual comprehensiveness in capturing the constructs.
usage. Based on their feedback, modifications were made to
the wording of some items and redundant or unclear
Gender differences in technology adoption have items were removed. This process resulted in a
been attributed to differences in cognitive styles, refined set of items for each construct. The refined
social influences, and self-efficacy (Venkatesh et al., set of items was pretested using a sample of 20
8 Information Development 0(0)

university students who were not part of the main Table 1. Demographic characteristics of the samples.
study. This step aimed to identify any potential Subjects (N = 232)
issues related to item comprehension, response
formats, and survey administration. Participants were Demographics Item Frequency Percentage
encouraged to provide feedback on any confusing or Gender Male 96 41.2%
ambiguous items, and their comments were taken Female 136 58.4%
into account to further refine the instrument. To Age 20 or younger 101 43.3%
account for the potential influence of gender and age 21 20 8.6%
on the relationships between the main constructs, 22 23 9.9%
these demographic variables were included in the 23 or older 88 37.8%
research model as control variables. The instrument
included single-item measures for both genders
(coded as 1 for male and 2 for female) and age (mea- sample size was deemed adequate based on the recom-
sured in years). All variables except for demographic mendations of Hair et al. (2011), who suggested a
information and frequency were assessed using a minimum sample size of 10 times the number of items
7-point Likert scale. Table A1 depicts all the measure- in the research model.
ment items for the constructs. Table 1 presents the demographic characteristics of
the participants (N = 232) in the study. The table is
organized into four age groups and two gender cat-
Subjects and data collection egories, displaying the frequency and percentage of
The target population for this study comprised univer- participants for each group. Among the participants,
sity students who were actual users of ChatGPT, an AI 41.2% (96 individuals) were male, and 58.4% (136
chatbot, in their learning process. A purposive sampling individuals) were female. In terms of age distribution,
technique was employed to select participants who had 43.3% (101 individuals) were 20 years old or younger,
some experience or familiarity with AI chatbots, ensur- 8.6% (20 individuals) were 21 years old, 9.9% (23
ing that they could provide meaningful responses to the individuals) were 22 years old, and 37.8% (88 indivi-
survey items. Data were collected through an online duals) were 23 years old or older.
survey administered to the selected university students. By following a rigorous subject selection and data col-
The survey link was shared through various channels, lection procedure, this study ensured that the collected
such as professors, online forums, and social media data were representative of the target population and suit-
groups related to the university community from the able for testing the research model in the context of
27th of March to the 8th of April, 2023. Some professors ChatGPT adoption among university students.
from various universities across South Korea assisted
the distribution and gathering. Professors encouraged
both undergraduate and postgraduate students to partici- Results
pate in the survey. The participation of the students was This study employed SmartPLS 3.3.9 software to analyze
entirely voluntary, and it is important to note that at the the collected data and evaluate the research model. It
time of the survey, the majority of students in a class had opted for partial least squares structural equation model-
experience using ChatGPT. In the survey invitation, the ing (PLS-SEM) as a primary analytical method for
purpose of the study was briefly explained, and partici- several reasons. First, PLS-SEM is a powerful technique
pants were assured that their responses would be for analyzing complex research models with multiple
anonymous and confidential. Upon clicking the survey variables, such as those in this study. This approach
link, participants were directed to an informed consent allows for the simultaneous assessment of the relation-
form outlining the study’s objectives, potential risks ships between variables and the examination of both
and benefits, and their rights as participants. After pro- direct and indirect effects (Hair et al., 2021). Second,
viding their consent, they were able to proceed with PLS-SEM is a variance-based method that is particularly
the survey. In this step, informed consent was obtained suitable for studies with smaller sample sizes and non-
from all participants. A total of 243 university students normal data distributions (Hair et al., 2011). In this
participated in the survey. After screening the responses study, we had a relatively small sample size, making
for incomplete or careless submissions, a final sample of PLS-SEM a more appropriate choice compared to
232 usable responses was retained for data analysis. This covariance-based SEM techniques, which typically
Jo: Decoding the ChatGPT mystery 9

require larger sample sizes for accurate results. Third, ChatGPT as beneficial for their study processes.
PLS-SEM is a predictive modeling approach, which is Similarly, Perceived intelligence (5.172 to 5.487) also
consistent with this study’s objective to understand the ranked high, showing that students regard ChatGPT as
determinants of ChatGPT usage among university a highly intelligent AI system. The usability score
students. The method allows us to not only explore the (5.039 to 5.371) indicates that most students found
relationships between variables but also identify the ChatGPT easy to use and navigate. Moreover, the atti-
key factors that contribute to the prediction of the depend- tude, subjective norms, and perceived behavioral
ent variable (Ringle et al., 2015). Finally, PLS-SEM is control constructs, with mean scores above the mid-
well-suited for exploratory research and theory develop- point, imply a generally positive disposition, social pres-
ment, as it can accommodate both reflective and forma- sure, and perceived control towards using ChatGPT.
tive measurement models (Hair et al., 2019). This study Interestingly, behavioral intention (4.353 to 5.073)
seeks to extend the theory of planned behavior by incorp- showed that students intend to use ChatGPT, albeit
orating additional factors, such as knowledge applica- with a bit of variation in their responses. However,
tion, perceived intelligence, and usability, making trust (3.409 to 3.677) and actual usage (3.203 to
PLS-SEM a suitable method for testing the relationships 3.724), scored notably below the midpoint of the
in our study model. scale. It implies that although students perceive
ChatGPT as a beneficial and intelligent system, they
Common method bias (CMB) have concerns about its trustworthiness, particularly
regarding information security. This concern might
This study employs three techniques to address potential explain the lower actual usage of ChatGPT, despite
CMB. Firstly, Harman’s one-factor test, a principal com- the high behavioral intention. In sum, these statistics
ponent analysis technique, was used. This test suggests show that while students acknowledge ChatGPT’s intel-
that if a single factor accounts for more than 50% of the ligence and potential, they harbor reservations, particu-
variance, CMB is present (Podsakoff et al., 2003). The larly about trust, which could affect their actual usage of
analysis revealed a single factor variance of 37.920%, the system.
which is less than the 50% threshold, indicating the The measurement model was assessed by examining
absence of CMB. Secondly, variance inflation factors the reliability, convergent validity, and discriminant val-
(VIFs) were analyzed. A VIF value exceeding 3.3 can idity of the constructs. Composite reliability (CR) and
signify CMB (Kock, 2015), but all constructs in this Cronbach’s alpha were used to evaluate the internal con-
study had VIFs below this threshold, further suggesting sistency reliability of the constructs. Factor loadings,
CMB is not a significant issue. Lastly, the comprehensive average variance extracted (AVE), and the correlation
confirmatory factor analysis marker technique proposed matrix were examined to evaluate convergent and dis-
by Williams et al. (2010) was employed. This technique criminant validity. The results indicated satisfactory reli-
uses a theoretically unrelated marker variable to estimate ability, as all constructs exhibited CR values above the
the extent of method variance. We chose a marker with recommended threshold of 0.7 and Cronbach’s alpha
low theoretical correlations to the main constructs, built values above 0.7, suggesting adequate internal consist-
a measurement model that included all main constructs ency (Hair et al., 2017). The AVE values for all con-
and the marker variable, and set conditions on the structs exceeded the recommended threshold of 0.5,
marker loadings, error variances, and marker’s variance. indicating sufficient convergent validity (Fornell and
This process generated an estimate of the method vari- Larcker, 1981). The factor loadings of all indicators
ance, which suggested that the influence of method vari- were above the recommended value of 0.7.
ance on the findings was minimal. These three methods, Discriminant validity was established, as the square
combined, suggest that this study effectively addresses root of AVE for each construct was greater than its
concerns of CMB. correlation with other constructs (Fornell and
Larcker, 1981). Table 3 shows the results of the dis-
criminant validity assessment.
Reliability and validity of the data Table 4 presents the Heterotrait-Monotrait (HTMT)
Table 2 shows descriptive statistics, reliability, and con- ratio values for the constructs in the study. The table
vergent validity. Knowledge application (ranging from displays the HTMT ratios between each pair of con-
4.651 to 5.250) scored above the scale mid-point, sug- structs, which are used to assess discriminant validity.
gesting that students generally perceive the use of Values less than 0.9 indicate adequate discriminant
10 Information Development 0(0)

Table 2. Reliability and convergent validity.

Construct Items Mean St Dev. Factor Loading Cronbach’s Alpha CR AVE
Knowledge Application KAP1 5.250 1.248 0.900 0.822 0.894 0.739
KAP2 5.065 1.235 0.896
KAP3 4.651 1.421 0.778
Perceived Intelligence PIE1 5.487 1.228 0.885 0.862 0.915 0.782
PIE2 5.172 1.331 0.878
PIE3 5.280 1.369 0.890
Usability USB1 5.371 1.290 0.822 0.732 0.845 0.646
USB2 5.272 1.358 0.837
USB3 5.039 1.388 0.750
Attitude ATT1 5.810 1.037 0.888 0.762 0.863 0.678
ATT2 6.004 1.048 0.844
ATT3 5.056 1.314 0.731
Subjective Norms SNO1 5.366 1.167 0.919 0.921 0.950 0.863
SNO2 5.151 1.283 0.928
SNO3 5.323 1.194 0.940
Perceived Behavioral Control PBC1 5.728 1.185 0.877 0.865 0.917 0.787
PBC2 5.586 1.243 0.892
PBC3 5.440 1.195 0.891
Trust TRU1 3.677 1.641 0.931 0.936 0.959 0.885
TRU2 3.591 1.584 0.955
TRU3 3.409 1.576 0.936
Behavioral Intention BIT1 5.073 1.497 0.918 0.885 0.929 0.813
BIT2 4.353 1.652 0.865
BIT3 5.039 1.445 0.921
Actual Usage ACU1 3.724 1.469 0.958 0.881 0.943 0.892
ACU2 3.203 1.678 0.931

Table 3. Fornell-Larcker scale results.

Constructs 1 2 3 4 5 6 7 8
1. Knowledge Application 0.860
2. Perceived Intelligence 0.677 0.884
3. Usability 0.395 0.392 0.804
4. Attitude 0.569 0.623 0.441 0.824
5. Subjective Norms 0.463 0.515 0.319 0.501 0.929
6. Perceived Behavioral Control 0.437 0.480 0.323 0.438 0.537 0.887
7. Trust 0.177 0.188 0.159 0.157 0.136 0.105 0.941
8. Behavioral Intention 0.542 0.542 0.348 0.572 0.699 0.477 0.307 0.902

Table 4. HTMT.
Constructs 1 2 3 4 5 6 7 8
1. Knowledge Application
2. Perceived Intelligence 0.795
3. Usability 0.494 0.470
4. Attitude 0.707 0.743 0.557
5. Subjective Norms 0.530 0.565 0.369 0.594
6. Perceived Behavioral Control 0.513 0.542 0.384 0.529 0.600
7. Trust 0.197 0.196 0.199 0.200 0.142 0.115
8. Behavioral Intention 0.638 0.609 0.397 0.688 0.769 0.534 0.336
Jo: Decoding the ChatGPT mystery 11

validity between the constructs, as suggested by previ- provided support for the majority of the proposed hypoth-
ous research (Henseler et al., 2015). eses and demonstrated the suitability of the research
model in explaining the factors influencing university
Structural model students’ adoption and usage of ChatGPT. Figure 2 dis-
plays the results of the structural model.
After validating the measurement model, the structural Overall, the theoretical model accounts for approxi-
model was assessed by examining the path coefficients, mately 38.8 percent of the variance in actual usage.
t-values, and R-squared values for the dependent vari- Table 5 describes the analysis results of SEM.
ables. Bootstrapping with 5000 resamples was per-
formed to obtain the t-values and test the significance
of the path coefficients. The results showed that most Post-hoc analysis
hypotheses were supported, with significant path coeffi- The study also explored potential curvilinear effects of per-
cients and t-values above the critical value of 1.96. The R2 ceived intelligence on users’ attitudes towards ChatGPT.
values for the dependent variables indicated the propor- Specifically, a post-hoc analysis was performed to assess
tion of variance explained by the model. The R2 values the possibility of an inverted U-shaped relationship, fol-
were moderate, suggesting that the model explained a lowing the rationale that an excessively high level of per-
considerable amount of variance in the dependent vari- ceived intelligence might be perceived as uncanny or
ables. In summary, the analysis using SmartPLS 3.3.9 intimidating, thus deterring users. This conjecture is akin

Figure 2. Analysis results (PLS Algorithm).

Table 5. Analysis of path coefficients.

H Cause Effect Coefficient T-value P-value Hypothesis
H1 Knowledge Application Attitude 0.229 3.004 0.003 Supported
H2a Perceived Intelligence Knowledge Application 0.676 16.039 0.000 Supported
H2b Perceived Intelligence Usability 0.400 7.342 0.000 Supported
H2c Perceived Intelligence Attitude 0.387 5.430 0.000 Supported
H2d Perceived Intelligence Trust 0.185 2.758 0.006 Supported
H3 Usability Attitude 0.197 3.480 0.001 Supported
H4 Attitude Behavioral Intention 0.256 4.687 0.000 Supported
H5 Subjective Norms Behavioral Intention 0.504 8.542 0.000 Supported
H6 Perceived Behavioral Control Behavioral Intention 0.074 1.139 0.255 Not Supported
H7 Trust Behavioral Intention 0.190 4.262 0.000 Supported
H8 Behavioral Intention Actual Use 0.601 14.056 0.000 Supported
12 Information Development 0(0)

to findings in the field of robotics, where entities that Furthermore, this outcome highlights the importance of
appear too lifelike can provoke discomfort and disengage- developing AI systems that can effectively synergize
ment, a phenomenon referred to as the “uncanny valley” with human users, fostering a more cooperative relation-
effect (Kim et al., 2019; Laakasuo et al., 2021; ship. This, in turn, can expedite the emergence of hybrid
Söderlund, 2022). intelligence systems, wherein human and AI capabilities
The post-hoc analysis showed an insignificant coeffi- mutually augment one another.
cient for the curvilinear effect of perceived intelligence The significant effects of perceived intelligence on
on attitude (β = -0.023, p = 0.433). This indicates a lack knowledge application, usability, attitude, and trust in
of support for an inverted U-shaped relationship the context of AI like ChatGPT, align well with previous
between perceived intelligence and attitude. However, research findings and theoretical expectations. Starting
this does not necessarily refute the uncanny valley with knowledge application, it makes intuitive sense
hypothesis in the context of AI interfaces like that if users perceive ChatGPT as intelligent, they
ChatGPT, as this theory primarily applies to physical would expect it to facilitate their understanding and appli-
and visually represented entities. The disembodied cation of complex knowledge (Sarker, 2022). The per-
nature of ChatGPT may attenuate any uncanny or ceived intelligence of ChatGPT would enhance its
intimidating effects of high perceived intelligence. perceived value as an effective educational tool, capable
of distilling and presenting complex information in a
useful and digestible way. The more intelligent the AI
Discussion system is perceived to be, the more likely it is to be
The findings of this study revealed several significant used to aid knowledge application. This aspect also sup-
relationships among the constructs, providing critical ports the concept of human-augmented AI, where human
insights into the factors influencing university stu- intelligence amplifies machine capabilities through con-
dents’ adoption of ChatGPT. tinuous interaction and learning (Jarrahi et al., 2022).
The observation of a positive effect of knowledge Regarding usability, perceived intelligence is likely to
application on attitude corresponds to existing research play a crucial role as well. The smarter the AI system is
underscoring the significance of perceived usefulness perceived to be, the more seamless and intuitive the
in technology adoption (De Cicco et al., 2022; user interaction is expected to be (Moussawi et al.,
Nadarzynski et al., 2019). This effect can be compre- 2021). The intelligence of the system should allow it to
hended by acknowledging that students in universities better understand the user’s requests, reducing confusion
are motivated by their aspiration to enhance their learning and enhancing the overall user experience. The impact of
and comprehension of concepts. Consequently, when an perceived intelligence on attitude aligns with the past
AI resource, such as ChatGPT, facilitates knowledge works (Balakrishnan and Dwivedi, 2021; Tussyadiah
application or reinforcement, it is perceived as valuable, and Park, 2018). Individuals’ attitudes towards a system
leading to a positive attitude towards the tool (Teo, 2009). are likely to be more positive if they perceive the
Students tend to develop favorable attitudes towards system to be intelligent. In this context, the perceived
instruments that enrich their learning experiences, assist intelligence of ChatGPT can contribute to attitudes
in understanding intricate topics, and bolster academic towards the system. Lastly, perceived intelligence likely
achievement. Linking these findings to human-AI collab- impacts trust as users often associate intelligence with
oration and hybrid intelligence offers additional insights. reliability and predictability (Castelfranchi and Falcone,
Persistent interaction between humans and AI can lead to 2010). The more intelligent an AI system is perceived
the mutual enhancement of intelligence (Jarrahi et al., to be, the more it can anticipate user needs, avoid mis-
2023, 2022). In this scenario, users’ ability to practically takes, and provide consistent output, all of which can
utilize the knowledge gained from AI (knowledge appli- foster user trust. Trust, built on perceived intelligence,
cation) favorably impacts their attitude towards AI. This can encourage users to accept AI’s recommendations
infers that users appreciate AI not merely for the raw and incorporate them into their decision-making
information it imparts but for its potential to aid in prac- process, thereby fostering a more effective human-AI
tical problem-solving circumstances, thereby reinforcing symbiosis (Jarrahi et al., 2022).
the significance of human-AI synergies. As users identify The positive association between usability and atti-
meaningful ways to apply the knowledge they acquire tude aligns with prior literature on the role of conveni-
from AI, they begin to view the AI as a collaborative ence and ease of access in AI adoption (Rafiq et al.,
partner, enhancing their overall attitude towards it. 2022). When a system is perceived as usable, it
Jo: Decoding the ChatGPT mystery 13

means that users can interact with it effectively, effi- Lai et al., 2022), confirming that the stronger the
ciently, and satisfactorily in a specified context of intention to use ChatGPT, the higher the likelihood
use (Petter et al., 2008). This enhances the user’s posi- of actual usage.
tive feelings or attitude toward the system, thus pro- Lastly, the inferred non-linearity of this relationship
moting system use. In the context of ChatGPT, the implies an optimal level of perceived intelligence that
findings suggest that when users perceive the system maximizes user attitude towards the AI. Despite this
as easy to navigate, understand, and interact with pattern not being statistically significant, a negative
(high usability), they are more likely to have a positive inverse U-shape relationship suggests that an increase
attitude toward it. This positive attitude then fosters in perceived intelligence initially improves the attitude
increased adoption and usage of ChatGPT. towards it, up to a certain point. Beyond this limit, any
The significant relationship between attitude and further increase in perceived intelligence may lead to a
behavioral Intention, as well as subjective norms and decline in positive attitudes, indicating the uncanny
behavioral intention, is consistent with the TPB (Ajzen, valley phenomenon (Kim et al., 2019; Laakasuo
1991). A positive attitude towards ChatGPT encourages et al., 2021; Söderlund, 2022). These results warrant
students to consider adopting the technology, as they per- further exploration into the complex impacts of per-
ceive it as beneficial and enjoyable to use. Additionally, ceived intelligence on user attitudes towards AI,
subjective norms, which represent the social influence including potential moderating influences and
of peers and important individuals, also impact varying effects across distinct user subgroups. This
student’’ intentions. When students observe their peers analysis emphasizes the critical importance of striking
using ChatGPT or receive recommendations from sig- a balance in perceived intelligence when developing
nificant others, they are more likely to develop a behav- and deploying AI interfaces in educational settings.
ioral intention to adopt the technology themselves. While an optimal degree of perceived intelligence
These results emphasize the importance of fostering a can promote user engagement and acceptance, an
positive attitude and leveraging social influence to excessively high level might risk creating a sense of
promote ChatGPT adoption among university students. alienation among users.
The non-significant relationship between perceived
behavioral control and behavioral intention contrasts Conclusion
with the past research (Chai et al., 2020; Chao, 2019;
Hansen et al., 2018), suggesting that other factors, such Theoretical contributions
as the ones identified in this study, might play a more sig- This study makes significant theoretical contributions
nificant role in shaping student’’ intentions to use It high- by providing a more comprehensive understanding of
lights the need to focus on other factors such as attitude, users’ perceptions of AI and its consequent impacts on
subjective norms, and trust, which have a more signifi- their interactions and relationships with this technol-
cant influence on adoption intentions in this context. ogy. One unique aspect of this research is its explor-
The significant relationship between trust and ation of the interplay between perceived intelligence
behavioral intention underscores the importance of and knowledge application. This dynamic, often over-
trust in technology adoption (Brill et al., 2019; Han looked or understudied in previous literature, sheds
and Yang, 2018; Nguyen et al., 2019). The finding light on a critical dimension of AI-user engagement.
suggests that as university students’ trust in the infor- Prior studies have noted the role of perceived intelli-
mation security of ChatGPT increases, so does their gence in fostering user engagement, with the under-
intention to use the system. This relationship may be lying assumption that smarter AI systems tend to be
due to the nature of the information often handled more engaging (Aw et al., 2022; Rafiq et al., 2022).
by ChatGPT, which can include sensitive data such However, this research pushes the boundary further
as personal details, academic data, and other confiden- by uncovering that perceived intelligence significantly
tial information. As students perceive ChatGPT as affects how users apply the knowledge they gain from
secure and trustworthy, they are more likely to AI in real-world situations. This finding moves the
intend to use it (McKnight et al., 2002). dialogue from a mere focus on user engagement to
Moreover, the significant positive effect of behav- the actual practical value derived from interactions
ioral intention on actual usage is in line with prior with AI. The study’s findings underscore the import-
research on technology adoption (A. Khan & ance of perceived intelligence not just in driving
Qudrat-Ullah, 2021; Agudo-Peregrina et al., 2014; user engagement, but more importantly, in enhancing
14 Information Development 0(0)

the practicality of the AI system in terms of knowl- interaction, which can pave the way for further
edge application. This has far-reaching implications explorations and investigations in this domain.
in understanding and assessing the value and effect- Finally, this study offers unique insights into the
iveness of AI systems, particularly in educational delicate equilibrium that characterizes the relationship
and other knowledge-intensive contexts. Rather than between perceived intelligence and user attitude. Prior
a mere tool for information provision, AI systems research has predominantly examined this relationship
are seen as potential partners in knowledge creation from a linear perspective (Balakrishnan et al., 2022;
and application, further highlighting the practical Balakrishnan and Dwivedi, 2021; Rafiq et al., 2022).
and collaborative role of AI in learning contexts. However, this research suggests the existence of an
Adding another layer to our theoretical understand- inverse U-shaped relationship, although statistically
ing, this study successfully integrates the constructs of insignificant. Despite its statistical insignificance, the
human-AI collaboration and hybrid intelligence suggested relationship could potentially provide a
(Jarrahi et al., 2023, 2022). Previous research has pro- new theoretical perspective to the body of AI percep-
vided substantial evidence supporting the significant tion literature, indicating a need for further exploration
role of perceived intelligence in shaping users’ atti- in this direction. This highlights that while a certain
tudes towards AI (Aw et al., 2022; Balakrishnan degree of perceived intelligence is beneficial for posi-
et al., 2022; Moussawi and Koufaris, 2019). tive user attitudes, an overly high perception may lead
However, these studies have not thoroughly explored to reduced positivity. It brings to light the notion of
how perceived intelligence could foster human-AI syn- ‘optimal perceived intelligence’ that can maximize
ergies, a gap this research endeavors to fill. In the user attitudes, opening new avenues for academic
context of human-AI collaboration, this study exploration. This finding challenges the prevailing
uncovers that as perceived intelligence increases, assumption that increased intelligence always corre-
users begin to view AI not merely as a tool but as a lates positively with user attitude and calls upon scho-
co-working entity or a collaborator. This shift in per- lars to delve deeper into understanding the effects of
ception is critical as it enhances the potential for the varying levels of perceived intelligence on user atti-
development of hybrid intelligence systems where tude, trust, usability, and knowledge application.
human and AI capabilities are not just complementary This finding also has implications for AI design and
but are integrated to achieve better results. This implementation, emphasizing the need for designers
research further emphasizes that perceived intelligence to consider this optimal level of perceived intelligence
can have profound effects on the nature and effective- in order to foster positive attitudes and successful
ness of human-AI collaborations. Therefore, it is user-AI interactions.
crucial for future research to investigate how perceived
intelligence impacts the dynamics of human-AI colla-
borations across different contexts, and how this per- Managerial implications
ception can be effectively managed to optimize The practical implications of this study are manifold and
collaboration and the overall performance of hybrid present actionable insights for several key stakeholders,
intelligence systems. including service providers, product manufacturers,
The paper puts forth an important theoretical impli- managers, developers, marketers, and users. The find-
cation concerning the effect of perceived intelligence ings particularly highlight the importance of understand-
on trust in AI. Prior studies have demonstrated that ing users’ perceptions of AI and its implications on their
trust is crucial for user acceptance and adoption of interaction with the technology.
AI (Siau and Yang, 2017). Nevertheless, this research Firstly, the study offers crucial insights for practi-
offers a more detailed comprehension by revealing tioners concerning not only the design of AI but also
that user trust is significantly influenced by the per- its application in knowledge utilization. The research
ceived intelligence of AI. This could potentially underscores a significant positive influence of knowl-
pave the way for further research exploring how edge application on attitudes towards AI. This infers
enhancing perceived intelligence can bolster trust that when AI systems are utilized in environments
and thus, increase user adoption and acceptance. In facilitating the application of acquired knowledge–
conclusion, this study emphasizes the need for a com- including educational, problem-solving, or profes-
prehensive and detailed comprehension of perceived sional development contexts–it engenders a more
intelligence and its diverse effects on user-AI favorable disposition towards the technology. These
Jo: Decoding the ChatGPT mystery 15

insights are invaluable for service providers and edu- human-AI collaboration and maximizing the potential
cational institutions. AI’s utility is augmented when of hybrid intelligence.
its applications coincide with real-world problems, Thirdly, the insights gleaned from this study offer
enabling users to apply the obtained knowledge prac- valuable implications for marketers aiming to effect-
tically (Sarker, 2022). Hence, providers, particularly ively communicate the benefits of AI technologies to
within the education sector or other knowledge- potential users. Understanding the balance between
intensive industries, should highlight the practical the perceived intelligence and user attitudes towards
applications of AI technologies in their marketing AI allows marketers to frame their AI products in a
and engagement strategies. The use of specific case manner that is most likely to elicit positive responses.
studies or examples where AI has been instrumental For instance, the campaign can emphasize its ability to
in augmenting human learning, decision-making, or provide tailored, intelligent feedback that enhances
problem-solving could delineate the real-world bene- students’ understanding of complex concepts, reinfor-
fits of these technologies (Mentzas et al., 2021). cing the utility of the AI tool in knowledge applica-
Such an approach would not only enhance user atti- tion. At the same time, it’s vital to highlight that the
tudes towards AI but could also stimulate increased AI tutor, despite its sophisticated intelligence, is
adoption and user engagement. This focus on knowl- designed to deliver an approachable and non-
edge application accentuates AI’s potential as an inter- intimidating user experience, thereby not overwhelm-
active learning instrument that stimulates active ing the users. The emphasis should be on AI’s ability
engagement and learning transfer, rather than a to augment human capacities, rather than replace
passive information source. Therefore, educators, them, reinforcing the potential of AI as a tool for
instructional designers, and AI developers should human-AI synergies. It is through this effective com-
aim to design environments that inspire knowledge munication that marketers can not only improve the
application, thereby enhancing the learning process acceptance of AI products but also foster a better
and fostering more positive user attitudes towards AI. understanding of the capabilities and potential of AI
The second practical implication arising from this technologies among users.
study is for managers and service providers, who Lastly, this research’s finding about the influence of
need to recognize the significance of nurturing subjective norms on students’ behavioral intention to
human-AI synergies. This study discovered that as per- use ChatGPT bears significant implications for educa-
ceived intelligence escalates, users increasingly view tional institutions. It highlights the need to foster a
AI as a collaborative partner, promoting the evolution learning culture that promotes the acceptance and
of hybrid intelligence systems. From a practical stand- use of AI chatbots like ChatGPT. Educational institu-
point, this could necessitate the structuring of work pro- tions can play a pivotal role in shaping these subject-
cedures to integrate AI capabilities, facilitating efficient ive norms and encouraging chatbot adoption. Faculty
collaboration between human workforce and AI members, for instance, can incorporate ChatGPT into
systems. For example, an AI system could be imple- their pedagogical approach by integrating it into
mented to provide employees with data-driven insights, course materials and assignments. Such integration
helping them make informed decisions (Ledro et al., can help students realize the practical value of AI chat-
2022). Alternatively, AI could automate routine tasks, bots in academic contexts, thereby encouraging their
allowing employees to focus on more complex, stra- use (Sun and Hoelscher, 2023). Moreover, administra-
tegic initiatives (Deranty and Corbin, 2022). This tive stakeholders can facilitate workshops, seminars,
process not only optimizes productivity but also culti- or informational sessions that highlight the benefits
vates a more positive disposition towards AI among of using AI chatbots for learning. These activities
the workforce. Such positive attitudes can further can demystify the technology, address potential appre-
encourage employees to explore innovative ways of hensions, and illustrate the practical benefits that can
leveraging AI capabilities, driving further benefits for be derived from these tools. By fostering an institu-
the organization. Consequently, managers and service tional environment that endorses the use of AI chat-
providers should consider such human-AI collabor- bots, educational institutions can strengthen the
ation when designing their service offerings and oper- subjective norms around chatbot use and influence
ational processes. They might also need to provide student behavioral intentions positively. This pro-
relevant training for employees to effectively use active approach can effectively drive the adoption of
AI tools and systems, further fostering successful AI tools like ChatGPT in learning settings, ultimately
16 Information Development 0(0)

enhancing the learning experience and outcomes for BEHAVIORAL CONTROL, trust, and actual usage
students. of AI chatbots like ChatGPT.

Declaration of conflicting interests

Limitation and future research
The author(s) declared no potential conflicts of interest with
This study has several limitations that offer opportun- respect to the research, authorship, and/or publication of
ities for future research. Firstly, our sample consisted this article.
solely of university students, which limits the general-
izability of the findings to other populations. Future Funding
research could explore the determinants of ChatGPT The author(s) received no financial support for the research,
adoption among other user groups, such as high authorship, and/or publication of this article.
school students, working professionals, or lifelong
learners, to better understand the factors that influence ORCID iD
AI chatbot adoption across different contexts. Hyeon Jo https://orcid.org/0000-0001-7442-4736
Secondly, while our study derives from the TPB and
concentrates on a limited set of factors, we acknow- References
ledge that there could be additional factors influencing Agudo-Peregrina ÁF, Hernández-García Á and Pascual-Miguel
ChatGPT adoption. These may include student’s FJ (2014) Behavioral intention, use behavior and the accept-
knowledge base, cognitive level, performance expect- ance of electronic learning systems: Differences between
ancy, hedonic motivation, and perceived risk. higher education and lifelong learning. Computers in
Human Behavior 34: 301–314.
Although our emphasis was primarily on the academic
Ajzen I (1991) The theory of planned behavior. Organizational
use of ChatGPT, it is worth noting that users might Behavior and Human Decision Processes 50(2): 179–211.
also interact with the AI system for enjoyment, pleas- Al-Emran M, AlQudah AA, Abbasi GA, et al. (2023)
ure, and entertainment. This hedonic motivation could Determinants of using AI-Based chatbots for knowledge
significantly influence user behavior and overall sharing: Evidence from PLS-SEM and fuzzy sets (fsQCA).
system adoption. Further, the study did not explore IEEE Transactions on Engineering Management: 1–15.
the mechanisms behind how ChatGPT processes Al-Emran M, Arpaci I and Salloum SA (2020) An empirical
information, which can be a critical factor impacting examination of continuous intention to use m-learning:
user perception and usage behavior. This extends to An integrated model. Education and Information
the consideration of the student’s ability to discern Technologies 25(4): 2899–2918.
between false or biased information and truthful infor- Al-Emran M and Mezhuyev V (2020) Examining the effect
mation when interacting with ChatGPT. Particularly of knowledge management factors on mobile learning
adoption through the use of importance-performance
in an educational setting, where the accuracy of infor-
map analysis (IPMA). In: Hassanien AE, Shaalan K
mation is crucial, this capability can have a significant and F M (eds) Tolba, Proceedings of the International
bearing on the level of trust and reliance on the AI Conference on Advanced Intelligent Systems and
system. In addition, individual variances, such as cul- Informatics 2019. Cham: Springer.
tural and demographic differences, could potentially Al-Emran M, Mezhuyev V and Kamaludin A (2018)
affect the perception and utilization of ChatGPT. Technology acceptance model in M-learning context: A
Our study did not address these factors. These unex- systematic review. Computers & Education 125: 389–412.
plored elements, coupled with the limitations identi- Al-Emran M and Teo T (2020) Do knowledge acquisition and
fied, provide substantial ground for future research. knowledge sharing really affect e-learning adoption? An
Thereby, they would contribute to a more holistic empirical study. Education and Information Technologies
understanding of ChatGPT usage and adoption. 25(3): 1983–1998.
Al-Sharafi MA, Al-Emran M, Iranmanesh M, et al. (2022)
Finally, this research did not consider the potential
Understanding the impact of knowledge management
moderating effects of individual differences, such as
factors on the sustainable use of AI-based chatbots for edu-
age, gender, or prior experience with AI chatbots, on cational purposes using a hybrid SEM-ANN approach.
the relationships between the constructs in our Interactive Learning Environments: 1–20.
model. Future research could investigate whether Alaiad A and Zhou L (2014) The determinants of home
these individual differences moderate the relationships healthcare robots adoption: An empirical investigation.
among knowledge application, perceived intelligence, International Journal Of Medical Informatics 83(11):
usability, attitude, subjective norms, PERCEIVED 825–840.
Jo: Decoding the ChatGPT mystery 17

Alavi M and Leidner DE (2001) Knowledge management and Buba AK, Ibrahim O and Shehzad HMF (2022) Behavioral
knowledge management systems: Conceptual foundations intention model for green information technology adop-
and research issues. MIS Quarterly 25(1): 107–136. tion in Nigerian manufacturing industries. Aslib Journal
AlHogail A (2018) Improving IoT technology adoption of Information Management 74(1): 158–180.
through improving consumer trust. Technologies 6(3): 64. Castelfranchi C and Falcone R (2010) Trust theory: A
Almaiah MA and Man M (2016) Empirical investigation to socio-cognitive and computational model. Hoboken,
explore factors that achieve high quality of mobile learn- NJ, USA: John Wiley & Sons.
ing system based on students’ perspectives. Engineering Chai CS, Wang X and Xu C (2020) An extended theory of
Science And Technology, An International Journal planned behavior for the modelling of Chinese second-
19(3): 1314–1320. ary school students’ intention to learn artificial intelli-
Ang T, Wei S and Arli D (2021) Social distancing behavior gence. Mathematics 8(11): 2089.
during COVID-19: A TPB perspective. Marketing Chang I-C, Shih Y-S and Kuo K-M (2022) Why would you
Intelligence & Planning 39(6): 809–824. use medical chatbots? Interview and survey. International
Aparicio M, Bacao F and Oliveira T (2017) Grit in the path Journal of Medical Informatics 165: 104827.
to e-learning success. Computers in Human Behavior Chao C-M (2019) Factors determining the behavioral inten-
66: 388–399. tion to use mobile learning: An application and extension
Ashfaq M, Yun J, Yu S, et al. (2020) I, Chatbot: Modeling of the UTAUT model. Frontiers in Psychology 10: 1652.
the determinants of users’ satisfaction and continuance Chung M, Ko E, Joung H, et al. (2020) Chatbot e-service
intention of AI-powered service agents. Telematics and and customer satisfaction regarding luxury brands.
Informatics 54: 101473. Journal of Business Research 117: 587–595.
Aw EC-X, Tan GW-H, Cham T-H, et al. (2022) Alexa, Cotton DRE, Cotton PA and Shipway JR (2023) Chatting
what’s on my shopping list? Transforming customer and cheating: Ensuring academic integrity in the era of
experience with digital voice assistants. Technological ChatGPT. Innovations in Education and Teaching
Forecasting and Social Change 180: 121711. International: 1–12.
Balakrishnan J, Abed SS and Jones P (2022) The role of Davis FD (1989) Perceived usefulness, perceived ease of
meta-UTAUT factors, perceived anthropomorphism, use, and user acceptance of information technology.
perceived intelligence, and social self-efficacy in MIS Quarterly 13(3): 319–340.
chatbot-based services? Technological Forecasting and De Cicco R, Iacobucci S, Aquino A, et al. (2022)
Social Change 180: 121692. Understanding Users’ Acceptance of Chatbots: An
Balakrishnan J and Dwivedi YK (2021) Conversational Extended TAM Approach 3–22Chatbot Research and
commerce: Entering the next stage of AI-powered Design. CONVERSATIONS 2021. Lecture Notes in
digital assistants. Annals of Operations Research: 1–35. Computer Science. vol. 13171 Cham: Springer.
Belanche D, Casaló LV and Flavián C (2019) Artificial Deranty J-P and Corbin T (2022) Artificial intelligence and
intelligence in FinTech: Understanding robo-advisors work: a critical review of recent research from the social
adoption among customers. Industrial Management & sciences. AI & Society.
Data Systems 119(7): 1411–1430. Devlin J, Chang M.-W, Lee K, et al. (2018). Bert:
Bhattacherjee A (2001) Understanding information systems Pre-training of deep bidirectional transformers for lan-
continuance: An expectation-confirmation model. MIS guage understanding. arXiv preprint arXiv:1810.04805.
Quarterly 25(3): 351–370. Dinev T and Hart P (2006) An extended privacy calculus
Borsci S, Malizia A, Schmettow M, et al. (2022) The chatbot model for e-commerce transactions. Information
usability scale: The design and pilot of a usability scale for Systems Research 17(1): 61–80.
interaction with AI-based conversational agents. Personal Dubey P and Sahu KK (2021) Students’ perceived benefits,
and Ubiquitous Computing 26(1): 95–119. adoption intention and satisfaction to technology-enhanced
Brandtzaeg PB and Følstad A (2017). Why people use chat- learning: Examining the relationships. Journal of Research
bots. Internet Science: 4th International Conference, in Innovative Teaching & Learning 14(3): 310–328.
INSCI 2017, Thessaloniki, Greece, November 22–24, Dwivedi YK, Kshetri N, Hughes L, et al. (2023) So what if
2017, Proceedings 4. ChatGPT wrote it?” Multidisciplinary perspectives on
Brill TM, Munoz L and Miller RJ (2019) Siri, Alexa, and opportunities, challenges and implications of generative
other digital assistants: A study of customer satisfaction conversational AI for research, practice and policy.
with artificial intelligence applications. Journal of International Journal of Information Management 71:
Marketing Management 35(15–16): 1401–1436. 102642.
Bruce E, Shurong Z, Ying D, et al. (2023) The effect of digital Farrokhnia M, Banihashem SK, Noroozi O, et al. (2023) A
marketing adoption on SMEs sustainable growth: SWOT analysis of ChatGPT: Implications for educa-
Empirical evidence from Ghana. Sustainability 15(6): tional practice and research. Innovations in Education
4760. https://www.mdpi.com/2071-1050/15/6/4760 and Teaching International: 1–15.
18 Information Development 0(0)

Følstad A, Nordheim CB and Bjørkli CA (2018) What Huang Y-C (2023) Integrated concepts of the UTAUT and
makes users trust a chatbot for customer service? An TPB in virtual reality behavioral intention. Journal of
exploratory interview study. In: Bodrunova SS (ed) Retailing and Consumer Services 70: 103127.
Internet Science. Cham: Springer, 194–208. Jarrahi MH, Askay D, Eshraghi A, et al. (2023) Artificial
Fornell C and Larcker DF (1981) Evaluating structural equa- intelligence and knowledge management: A partnership
tion models with unobservable variables and measure- between human and AI. Business Horizons 66(1): 87–99.
ment error. Journal of Marketing Research 18(1): 39–50. Jarrahi MH, Lutz C and Newlands G (2022) Artificial intel-
Gatzioufa P and Saprikis V (2022) A literature review on users’ ligence, human intelligence and hybrid intelligence
behavioral intention toward chatbots’ adoption. Applied based on mutual augmentation. Big Data & Society
Computing and Informatics: ahead-of-print(ahead-of- 9(2): 20539517221142824.
print). Johnson SB, King AJ, Warner EL, et al. (2023) Using
Gefen D, Karahanna E and Straub DW (2003) Trust and ChatGPT to evaluate cancer myths and misconceptions:
TAM in online shopping: An integrated model. MIS Artificial intelligence and cancer information. JNCI
Quarterly 27(1): 51–90. Cancer Spectrum 7(2): pkad015.
Hair J, Hollingsworth CL, Randolph AB, et al. (2017) An Joseph M (2023). ChatGPT history gone - how to retrieve
updated and expanded assessment of PLS-SEM in infor- the conversation history. STEALTH OPTIONAL.
mation systems research. Industrial Management & Retrieved August 28 from https://stealthoptional.com/
Data Systems 117(3): 442–458. how-to/chatgpt-history-gone-retrieve-conversation/
Hair JF, Hult GTM, Ringle CM, et al. (2021) A primer on Kai-ming Au A and Enderwick P (2000). A cognitive model
partial least squares structural equation modeling on attitude towards technology adoption. Journal of
(PLS-SEM). Thousand Oaks, CA: Sage Publications. Managerial Psychology 15(4): 266–282.
Hair JF, Ringle CM and Sarstedt M (2011) PLS-SEM: Khan RA and Qudrat-Ullah H (2021) Technology adoption
Indeed a silver bullet. Journal of Marketing Theory theories and models. In Adoption of LMS in Higher
and Practice 19(2): 139–152. Educational Institutions of the Middle East. Cham,
Hair JF, Risher JJ, Sarstedt M, et al. (2019) When to use and Zug: Springer International Publishing, 27–48.
how to report the results of PLS-SEM. European Kim SY, Schmitt BH and Thalmann NM (2019) Eliza in the
Business Review 31(1): 2–24. uncanny valley: Anthropomorphizing consumer robots
Han S and Yang H (2018) Understanding adoption of intel- increases their perceived warmth but decreases liking.
ligent personal assistants: A parasocial relationship per- Marketing Letters 30: 1–12.
spective. Industrial Management & Data Systems Kock N. (2015). WarpPLS 5.0 user manual. Laredo, TX:
118(3): 618–636. ScriptWarp Systems.
Hansen JM, Saridakis G and Benson V (2018) Risk, trust, Laakasuo M, Palomäki J and Köbis N (2021) Moral
and the interaction of perceived ease of use and behav- uncanny valley: A robot’s appearance moderates how
ioral control in predicting consumers’ use of social its decisions are judged. International Journal of
media for transactions. Computers in Human Behavior Social Robotics 13(7): 1679–1688.
80: 197–206. Lai Y, Saab N and Admiraal W (2022) University students’
Henseler J, Ringle CM and Sarstedt M (2015) A new criter- use of mobile technology in self-directed language learn-
ion for assessing discriminant validity in variance-based ing: Using the integrative model of behavior prediction.
structural equation modeling. Journal Of The Academy Computers & Education 179: 104413.
Of Marketing Science 43: 115–135. Laufer RS and Wolfe M (1977) Privacy as a concept and a
Hew KF, Huang W, Du J, et al. (2021). Using chatbots in social issue: A multidimensional developmental theory.
flipped learning online sessions: perceived usefulness Journal of Social Issues 33(3): 22–42.
and ease of use. In: Li R, Cheung SKS, Iwasaki C, et al Lebeuf C, Storey M-A and Zagalsky A (2017) Software
(eds) Blended Learning: Re-thinking and Re-defining bots. IEEE Software 35(1): 18–23.
the Learning Process. Cham: Springer, 164–175. Ledro C, Nosella A and Vinelli A (2022) Artificial intelli-
Ho SM, Ocasio-Velázquez M and Booth C (2017) Trust or gence in customer relationship management: Literature
consequences? Causal effects of perceived risk and sub- review and future research directions. Journal of
jective norms on cloud technology adoption. Computers Business & Industrial Marketing 37(13): 48–63.
& Security 70: 581–595. Lin D, Crabtree J, Dillo I, et al. (2020) The TRUST princi-
Hooda A, Gupta P, Jeyaraj A, et al. (2022) The effects of ples for digital repositories. Scientific Data 7(1): 144.
trust on behavioral intention and use behavior within Lo C K (2023) What is the impact of ChatGPT on education?
e-government contexts. International Journal of A rapid review of the literature. Education Sciences 13(4):
Information Management 67: 102553. 410. https://www.mdpi.com/2227-7102/13/4/410
Huang M-H and Rust RT (2018) Artificial intelligence in Lund BD, Wang T, Mannuru NR, et al. (2023) ChatGPT and a
service. Journal Of Service Research 21(2): 155–172. new academic reality: Artificial Intelligence-written
Jo: Decoding the ChatGPT mystery 19

research papers and the ethics of the large language models Nguyen QN, Ta A and Prybutok V (2019) An integrated
in scholarly publishing. Journal of the Association for model of voice-user interface continuance intention:
Information Science and Technology 74(5): 570–581. The gender effect. International Journal of Human–
McGinn C, Cullinan MF, Otubela M, et al. (2019) Design Computer Interaction 35(15): 1362–1377.
of a terrain adaptive wheeled robot for human-orientated Nguyen TH (2021). 5 Impactful Technologies From the
environments. Autonomous Robots 43: 63–78. Gartner Emerging Technologies and Trends Impact
McKnight D. (2005). Trust in Information Technology. In Radar for 2022. Gartner. https://www.gartner.com/en/
(pp. 329–331). articles/5-impactful-technologies-from-the-gartner-emerging-
McKnight DH, Choudhury V and Kacmar C (2002) The technologies-and-trends-impact-radar-for-2022
impact of initial consumer trust on intentions to transact Pal D, Babakerkhell MD and Zhang X (2021) Exploring the
with a web site: A trust building model. The Journal of Determinants of Users’ Continuance Usage Intention of
Strategic Information Systems 11(3-4): 297–323. Smart Voice Assistants. Ieee Access.
McLean G and Osei-Frimpong K (2019) Hey Alexa… Patil P, Tamilmani K, Rana NP, et al. (2020) Understanding
examine the variables influencing the use of artificial consumer adoption of mobile payment in India: Extending
intelligent in-home voice assistants. Computers in meta-UTAUT model with personal innovativeness,
Human Behavior 99: 28–37. anxiety, trust, and grievance redressal. International
Mentzas G, Lepenioti K, Bousdekis A, et al. (2021). Journal of Information Management 54: 102144.
Data-driven collaborative human-AI decision making. In: Petisca S, Dias J and Paiva A (2015) More social and emo-
Dennehy D, Griva A, Pouloudi N, et al (eds) tional behaviour may lead to poorer perceptions of a
Responsible AI and Analytics for an Ethical and social robot. Social Robotics: 7th International
Inclusive Digitized Society. Cham: Springer, 120–131. Conference, ICSR 2015, Paris, France, October 26-30,
Mishra A, Shukla A and Sharma SK (2022) Psychological 2015, Proceedings 7.
determinants of users’ adoption and word-of-mouth Petter S, DeLone W and McLean E (2008) Measuring infor-
recommendations of smart voice assistants. International mation systems success: Models, dimensions, measures,
Journal of Information Management 67: 102413. and interrelationships. European Journal of Information
Mohd Rahim NI, Iahad N, Yusof AF, et al. (2022). Systems 17(3): 236–263.
AI-Based Chatbots adoption model for higher-education Pillai R and Sivathanu B (2020) Adoption of AI-based
institutions: A hybrid PLS-SEM-neural network model- chatbots for hospitality and tourism. International
ling approach. Sustainability 14(19): 12726. Journal of Contemporary Hospitality Management
Mokhtar SSM and Salimon MG (2022). SMEs’ adoption of 32(10): 3199–3226.
artificial intelligence-chatbots for marketing communi- Podsakoff PM, MacKenzie M, Scott B, et al. (2003)
cation: A conceptual framework for an emerging Common method biases in behavioral research: A crit-
economy. In Marketing Communications and Brand ical review of the literature and recommended remedies.
Development in Emerging Markets Volume II: Insights Journal of Applied Psychology 885(879): 10–1037.
for a Changing World. Cham: Springer. 25–53. Rachmawati IK, Bukhori M, Nuryanti F, et al. (2020)
Morris, Michael G and Viswanath Venkatesh. (2000). Collaboration Technology Acceptance Model, subject-
Age differences in technology adoption decisions: ive norms and personal innovations on buying interest
Implications for a changing work force. Personnel online. International Journal of Innovative Science and
psychology, 53(2): 375–403. Research Technology 5(11): 115–122.
Moussawi S and Koufaris M (2019) Perceived intelligence Radford A, Wu J, Child R, et al. (2019) Language models are
and perceived anthropomorphism of personal intelligent unsupervised multitask learners. OpenAI Blog 1(8): 9.
agents: Scale development and validation. Rafiq F, Dogra N, Adil M, et al. (2022) Examining consu-
Moussawi S, Koufaris M and Benbunan-Fich R (2021) mer’s intention to adopt AI-chatbots in tourism using
How perceptions of intelligence and anthropomorphism partial least squares structural equation modeling
affect adoption of personal intelligent agents. Electronic method. Mathematics 10(13): 2190.
Markets 31(2): 343–364. Reeves B and Nass C (1996) The media equation: How
Nadarzynski T, Miles O, Cowie A, et al. (2019) people treat computers, television, and new media like
Acceptability of artificial intelligence (AI)-led chatbot real people. Cambridge, UK, 10, 236605.
services in healthcare: A mixed-methods study. Digital Ringle CM, Wende S and Becker J.-M (2015) “SmartPLS
Health 5: 2055207619871808. 3. Boenningstedt: SmartPLS GmbH, 31 2015”, Available
Nguyen DM, Chiu Y-TH and Le HD (2021). Determinants online: http://www.smartpls.com (accessed on 20 October
of continuance Intention towards Banks’ chatbot ser- 2020).
vices in Vietnam: A necessity for sustainable develop- Sandu N and Gide E (2019) Adoption of AI-Chatbots
ment. Sustainability 13(14): 7625. https://www.mdpi. to enhance student learning experience in higher educa-
com/2071-1050/13/14/7625 tion in India. 2019 18th International Conference on
20 Information Development 0(0)

Information Technology Based Higher Education and International Conference in Jönköping, Sweden, January
Training (ITHET). 24-26, 2018.
Sarker IH (2022) AI-Based modeling: Techniques, applica- Venkatesh V and Davis FD (2000) A theoretical extension
tions and research issues towards automation, intelligent of the technology acceptance model: Four longitudinal
and smart systems. SN Computer Science 3(2): 158. field studies. Management Science 46(2): 186–204.
Sciarelli M, Prisco A, Gheith MH, et al. (2022) Factors affect- Venkatesh V, Morris MG, Davis GB, et al. (2003) User
ing the adoption of blockchain technology in innovative acceptance of information technology: Toward a
Italian companies: An extended TAM approach. Journal unified view. MIS Quarterly 27(3): 425–478.
of Strategy and Management 15(3): 495–507. Wang SW, Ngamsiriudom W and Hsieh C-H (2015) Trust
Siau KL and Yang Y (2017) Impact of artificial intelligence, disposition, trust antecedents, trust, and behavioral
robotics, and machine learning on sales and marketing. intention. The Service Industries Journal 35(10):
Slade EL, Dwivedi YK, Piercy NC, et al. (2015) Modeling con- 555–572.
sumers’ adoption intentions of remote mobile payments in the Williams LJ, Hartman N and Cavazotte F (2010) Method
United Kingdom: Extending UTAUT with innovativeness, variance and marker variables: A review and comprehen-
risk, and trust. Psychology & Marketing 32(8): 860–873. sive CFA marker technique. Organizational Research
Söderlund M (2022) Service robots with (perceived) theory Methods 13(3): 477–514.
of mind: An examination of humans’ reactions. Journal Yan C, Siddik AB, Akter N, et al. (2023) Factors influencing
of Retailing and Consumer Services 67: 102999. the adoption intention of using mobile financial service
Song HG and Jo H (2023) Understanding the continuance during the COVID-19 pandemic: The role of FinTech.
intention of omnichannel: Combining TAM and TPB. Environmental Science and Pollution Research 30(22):
Sustainability 15(4): 3039. 61271–61289.
Study.com (2023) Productive Teaching Tool or Innovative Yoon J and Yu H (2022) Impact of customer experience on
Cheating? Retrieved April 28 from https://study.com/ attitude and utilization intention of a restaurant-menu
resources/perceptions-of-chatgpt-in-schools curation chatbot service. Journal of Hospitality and
Sullivan M, Kelly A and McLaughlan P (2023) ChatGPT in Tourism Technology 13(3): 527–541.
higher education: Considerations for academic integrity Zhang JJY, Følstad A and Bjørkli CA (2023) Organizational
and student learning. Journal of Applied Learning and factors affecting successful implementation of chatbots for
Teaching 6(1)): 1–10. customer service. Journal of Internet Commerce 22(1):
Sun GH and Hoelscher SH (2023) The ChatGPT storm and 122–156.
what faculty can do. Nurse Educator 48(3): 119–124. Zorz Z (2023). A bug revealed ChatGPT users’ chat
Sun S, Law R and Schuckert M (2020) Mediating effects of atti- history, personal and billing data. Help Net Security.
tude, subjective norms and perceived behavioural control for Retrieved April 20 from https://www.helpnetsecurity.
mobile payment-based hotel reservations. International com/2023/03/27/chatgpt-data-leak/
Journal of Hospitality Management 84: 102331.
Sundar SS, Bellur S, Oh J, et al. (2014) User experience of
on-screen interaction techniques: An experimental About the author
investigation of clicking, sliding, zooming, hovering,
dragging, and flipping. Human–Computer Interaction Hyeon Jo received his BS, MS, and PhD degrees from the
29(2): 109–152. Korea Advanced Institute of Science and Technology
Teo T (2009) Modelling technology acceptance in educa- (KAIST) in 2004, 2006, and 2012, respectively. His current
tion: A study of pre-service teachers. Computers & affiliation is with the HJ Institute of Technology and
Education 52(2): 302–312. Management, which conducts research related to emerging
Tlili A, Shehata B, Adarkwah MA, et al. (2023) What if the technology, business, and human behavior. His research inter-
devil is my guardian angel: ChatGPT as a case study of ests are emerging technology, information technology, and
using chatbots in education. Smart Learning Environments human-computer interaction. He has published in Journal of
10(1): 1–24.
Information Science, Journal of Business and Industrial
Tsai H-T, Chien J-L and Tsai M-T (2014) The influences of
Marketing, IT and Tourism, Journal of the Knowledge
system usability and user satisfaction on continued inter-
net banking services usage intention: Empirical evidence Economy, International Journal of Human-Computer
from Taiwan. Electronic Commerce Research 14(2): Interaction, Technology Analysis & Strategic Management,
137–169. Frontiers in Psychology, Heliyon, Humanities and Social
Tussyadiah IP and Park S (2018) Consumer evaluation of Sciences Communications, PLOS ONE, Scientific Reports,
hotel service robots. Information and Communication Asia Pacific Journal of Education, Asia Pacific Viewpoint,
Technologies in Tourism 2018: Proceedings of the Sustainability among others.
Jo: Decoding the ChatGPT mystery 21

Appendix

Table A1. List of constructs and items.

Construct Items Mean Source
Knowledge KAP1 ChatGPT provides me with instant access to various Al-Sharafi et al. (2022)
Application types of knowledge.
KAP2 ChatGPT allows me to integrate different types of
knowledge.
KAP3 ChatGPT helps me better master the knowledge
I learn in university.
Perceived PIE1 I feel that ChatGPT is competent. Rafiq et al. (2022)
Intelligence PIE2 I feel that ChatGPT is knowledgeable.
PIE3 I feel that ChatGPT is intelligent.
Usability USB1 The provided contents on ChatGPT are easily Rafiq et al. (2022); Borsci et al.
accessible. (2022); Almaiah and Man (2016)
USB2 The provided contents on ChatGPT are easily
downloadable.
USB3 The provided contents on ChatGPT are easily
retrievable.
Attitude ATT1 ChatGPT is useful. Ajzen (1991)
ATT2 ChatGPT is valuable.
ATT3 ChatGPT is beneficial.
Subjective Norms SNO1 Most people who are important to me think it is Ajzen (1991)
okay for me to use ChatGPT.
SNO2 Most people who are important to me support that
I use ChatGPT.
SNO3 Most people who are important to me understand
that I use ChatGPT.
Perceived PBC1 I am capable of using ChatGPT. Ajzen (1991)
Behavioral PBC2 I am confident that if I want to, I can use ChatGPT.
Control PBC3 I have enough resources, time, and opportunities to
use ChatGPT.
Trust TRU1 When using ChatGPT, I trust that my personal Nguyen et al. (2019)
information will not be used for any other
purpose.
TRU2 When using ChatGPT, I believe that my personal
information is protected.
TRU3 When using ChatGPT, I’m confident that my
personal information is secure.
Behavioral BIT1 I’d like to continue to use ChatGPT. Ajzen (1991)
Intention BIT2 I’ll continue to use the shared ChatGPT a lot in the
future.
BIT3 I plan to use ChatGPT in the future.
Actual Use ACU1 How frequently do you use ChatGPT? Aparicio et al. (2017); McLean and
ACU2 At present, I consider myself to be a frequent user of Osei-Frimpong (2019)
ChatGPT.