Computers in Human Behavior: Artificial Humans 2 (2024) 100027

Contents lists available at ScienceDirect

Computers in Human Behavior: Artificial Humans

journal homepage: www.journals.elsevier.com/computers-in-human-behavior-artificial-humans

ChatGPT in education: A blessing or a curse? A qualitative study exploring

early adopters’ utilization and perceptions
Reza Hadi Mogavi a, b, Chao Deng c, Justin Juho Kim b, Pengyuan Zhou d, Young D. Kwon e,
Ahmed Hosny Saleh Metwally a, f, Ahmed Tlili f, Simone Bassanelli g, Antonio Bucchiarone g,
Sujit Gujar h, Lennart E. Nacke a, Pan Hui b, i, j, *
a
University of Waterloo, Canada
b
The Hong Kong University of Science and Technology, Hong Kong, China
c
Accessible Meta Group, United States
d
University of Science and Technology of China, China
e
University of Cambridge, United Kingdom
f
Smart Learning Institute of Beijing Normal University, China
g
Fondazione Bruno Kessler, Italy
h
International Institute of Information Technology (Hyderabad), India
i
The Hong Kong University of Science and Technology (Guangzhou), China
j
University of Helsinki, Finland

A R T I C L E I N F O A B S T R A C T

Keywords: To foster the development of pedagogically potent and ethically sound AI-integrated learning landscapes, it is
Artificial intelligence (AI) pivotal to critically explore the perceptions and experiences of the users immersed in these contexts. In this study,
Generative AI we perform a thorough qualitative content analysis across four key social media platforms. Our goal is to un­
ChatGPT
derstand the user experience (UX) and views of early adopters of ChatGPT across different educational sectors.
Education
Human-computer interaction (HCI),
The results of our research show that ChatGPT is most commonly used in the domains of higher education, K-12
Early adopters education, and practical skills training. In social media dialogues, the topics most frequently associated with
Social media ChatGPT are productivity, efficiency, and ethics. Early adopters’ attitudes towards ChatGPT are multifaceted. On
Qualitative research one hand, some users view it as a transformative tool capable of amplifying student self-efficacy and learning
motivation. On the other hand, there is a degree of apprehension among concerned users. They worry about a
potential overdependence on the AI system, which they fear might encourage superficial learning habits and
erode students’ social and critical thinking skills. This dichotomy of opinions underscores the complexity of
Human-AI Interaction in educational contexts. Our investigation adds depth to this ongoing discourse, providing
crowd-sourced insights for educators and learners who are considering incorporating ChatGPT or similar
generative AI tools into their pedagogical strategies.

1. Introduction arguably the revolutionary impact on the educational sector (Seldon &
Abidoye, 2018; Zhai et al., 2021).
The twenty-first century is characterized by a rapid proliferation of This impact is best exemplified by the introduction of AI chatbots in
Artificial Intelligence (AI) technologies, leading to profound trans­ education, which provide instantaneous responses to students’ inquiries,
formations across various aspects of our daily lives, a theme explored in thus serving as readily accessible learning resources at any time and
numerous studies (e.g., see Wang et al., 2019; Aktan, Turhan, & Dolu, location (Clarizia, Colace, Lombardi, Pascale, & Santaniello, 2018;
2022; Wang & Lin, 2023; Ahn & Chen, 2020; Zhai et al., 2021; Hoi, Hwang & Chang, 2021; Kasthuri & Balaji, 2021). Concurrently, the use
2023). Among these transformative changes, the most significant is of cutting-edge machine learning algorithms to analyze student data has

* Corresponding author. The Hong Kong University of Science and Technology, Hong Kong, China.
E-mail addresses: [email protected] (R. Hadi Mogavi), [email protected] (C. Deng), [email protected] (J. Juho Kim), [email protected]
(P. Zhou), [email protected] (Y. D. Kwon), [email protected] (A. Hosny Saleh Metwally), [email protected] (A. Tlili), [email protected]
(S. Bassanelli), [email protected] (A. Bucchiarone), [email protected] (S. Gujar), [email protected] (L.E. Nacke), [email protected] (P. Hui).

https://doi.org/10.1016/j.chbah.2023.100027
Received 4 July 2023; Received in revised form 14 October 2023; Accepted 24 October 2023
Available online 20 November 2023
2949-8821/© 2023 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

empowered educators to pinpoint areas where individual students steps to fill this empirical research gap systematically by conducting a
require additional support, thereby enhancing the efficacy of teaching social media content analysis (SMCA) of four prominent social media
strategies (Zhang, 2022). Furthermore, the advent of AI-based learning platforms: Twitter,2 Reddit,3 YouTube,4 and LinkedIn.5
systems has paved the way for adaptive learning. These systems, SMCA is a popular research approach in the field of Human-
designed to adjust to each student’s learning pace and style, provide Computer Interaction (HCI) that allows researchers to capture peo­
bespoke educational content. Such personalization has demonstrated its ple’s reflective responses (Hadi Mogavi et al., 2022; Smith et al., 1992)
potential in improving learning outcomes and engagement levels (Liu to an emerging technology on a demographically large scale, which can
et al., 2020; Mogavi, Ma, & Hui, 2021; Xia et al., 2019). be otherwise quite difficult to establish through interviews, surveys, or
The Chat Generative Pre-trained Transformer (ChatGPT)1 is an other intrusive research methods (Blandford, Furniss, & Makri, 2016;
advanced chatbot technology that utilizes Natural Language Processing Hadi Mogavi et al., 2022; Sinha, Shibata, Patel, & Sternchos, 2021).
(NLP) to generate coherent and contextually-relevant responses to a Research Questions. More specifically, our research in this paper is
variety of human inquiries. In accordance with formal definitions, a guided by two main research questions:
chatbot is a “computer program designed to conduct a conversation with
a user through text or speech” (Chetlen, Artrip, Drury, Arbaiza, & RQ1: What are the predominant educational applications of
Moore, 2019). The ability of ChatGPT to create conversational and ChatGPT as reported by early adopters?
engaging learning environments, customized to individual student RQ2: What are the initial perceptions of adopters within the edu­
needs, underscores its potential for application in educational contexts. cation sector towards ChatGPT, and how do they evaluate its impact
This potential is increasingly recognized, as evidenced by an exponential on the educational process? (Is ChatGPT a blessing or a curse?)
growth in global interest in recent months, a trend observable in
worldwide Google search statistics (see Fig. 1). In our research, we consider the perspectives of various early adopter
This surge in interest highlights the promising prospects of ChatGPT groups, including researchers, educators, learners, parents, and general
for the development of advanced AI-powered educational systems. users (unspecified roles), all of whom could potentially play a critical role
However, it simultaneously underscores the need for a discerning in improving our understanding of the development, implementation,
approach when introducing this technology into educational settings. and use of educational technology (Cheng & Tsai, 2014; Hadi Mogavi
The caution advised in its usage stems from potential risks that have et al., 2021; Tlili et al., 2022, 2023). We posit that such a comprehensive
been identified in existing research (Sok, 2023). As such, the broadening investigation can inform the inclusive and responsible advancement of
interest in ChatGPT should be tempered by a careful evaluation of these ChatGPT in education, especially considering its nascent stage of
potential risks. adoption and development. In what follows, we outline some of the
Why Turn to Social Media for Insights? Since its official rollout, primary findings of our research:
ChatGPT has rapidly accumulated a significant user base (see Rubya). ChatGPT in Education. Utilizing an inductive bottom-up coding
Notably, social media has emerged as a primary venue for users to share approach, this study scrutinized approximately 150,000 words con­
their experiences and viewpoints regarding this AI tool (Tlili et al., tained in 6000 text samples to ascertain the prevalent applications of
2023). These publicly articulated opinions on social media platforms can ChatGPT across various educational contexts. A noteworthy adoption of
yield valuable insights into the usage of ChatGPT (Tlili et al., 2023). This ChatGPT was observed in higher education (24.18%), K-12 education
data enables researchers to delineate the primary educational applica­ (22.09%), and practical skills learning (15.28%). Additionally, ChatGPT
tions of ChatGPT and to assess its potential impacts, both positive and has found significant use in professional development (9.61%), entre­
negative, at an incipient stage. preneurship (7.18%), lifelong learning (4.64%), and popular science
With this background in mind, we are motivated to take the first education (3.25%). A total of 13.77% of the instances pertained to un­
classified learning areas.
ChatGPT’s versatility of use was demonstrated in its wide array of
applications including, but not limited to, assisting in essay and paper
composition, guiding learners through homework assignments/projects
and test preparation, aiding in language acquisition, fostering career
progression, honing communication strategies, improving digital liter­
acy, developing management competencies, facilitating benchmarking
and brainstorming sessions, supporting decision-making processes,
providing microlearning experiences, and simplifying complex concepts
in popular science education. Further details are elaborated in Section 4.
ChatGPT: A Double-Edged Sword. Discussions about ChatGPT on
social media platforms primarily focus on three themes: productivity
(73.19%), efficiency (63.02%), and ethics (48.51%).6 Our research sug­
gests that ChatGPT can have both positive and negative implications for
productivity, efficiency, and ethics.
Pioneer users of ChatGPT in educational contexts have expressed
varied perceptions. Specifically, 30.62% reported solely positive expe­
riences, 17.15% offered only negative feedback, 37.34% expressed
mixed feelings, and 14.89% maintained a neutral stance.
On one hand, ChatGPT is seen as a helpful tool that enhances

Fig. 1. The weekly search interest in ChatGPT on Google has consistently

increased since its inception, according to Statista (Bianchi, 2023). 2
https://twitter.com/
3
https://www.reddit.com/
4
https://www.youtube.com/
5
https://www.linkedin.com/
6
Readers should note that these percentages are not mutually exclusive;
1
https://chat.openai.com/ some user quotes may fall into multiple categories.

2
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

students’ self-efficacy and learning motivation. This is particularly Vaswani et al. (2017) has been highly influential in the field, serving as
beneficial for students with special needs and/or everyone else who the foundation for many large language models such as the GPT family
might struggle with traditional teaching methodologies. The conversa­ (Wolf et al., 2020).
tional style of ChatGPT enriches the learning experience. Its capacity to Transformers use a self-attention mechanism to process input tokens
provide immediate feedback and personalized learning opportunities in parallel, as opposed to sequentially, which is a key difference from
fosters a deeper connection between students and the subject matter, recurrent neural networks (RNNs) (Vaswani et al., 2017; Wolf et al.,
leading to enhanced knowledge retention. 2020). This allows transformers to be more efficient in processing large
Furthermore, the round-the-clock availability of ChatGPT is a boon amounts of text data and capturing long-range dependencies between
for students with demanding schedules, ensuring they do not miss out on words. The self-attention mechanism enables the model to better un­
learning opportunities. This is especially important for students from derstand relationships between words regardless of their position in the
socioeconomically disadvantaged backgrounds who may have commit­ text, which improves performance on NLP tasks (Wolf et al., 2020).
ments that conflict with traditional learning schedules. Language models have greatly benefitted from the use of pre-training
On the other hand, some concerns exist. There are apprehensions techniques (Dong et al., 2019). These techniques involve initially
that students might rely excessively on AI systems, leading to superficial training a model on an extensive dataset before fine-tuning it for a
learning, reduced social skills, and impaired communication abilities. It specific task. Pre-trained models such as BERT (Bidirectional Encoder
is also feared that it might hinder the development of key psychological Representations from Transformers) (Devlin, Chang, Lee, & Toutanova,
traits and discourage exploration of alternative viewpoints. 2019), BART (Bidirectional and Auto-Regressive Transformers) (Lewis
In addition, the deployment of ChatGPT in education raises vital et al., 2019), XLNet (Yang et al., 2019), and LLaMa from Facebook/Meta
concerns about data privacy and ethics. Further details are elaborated in (Touvron et al., 2023) have showcased significant improvements in
Section 4. performance across various natural language processing tasks.
Contributions. In summation, this study offers several primary Most large language models today possess the capability to acquire
contributions to the field: novel tasks with minimal examples, even if they are not directly asso­
ciated with their initial training (Beltagy, Cohan, Logan IV, Min, &
⋅ We have developed a comprehensive resource that catalogues the Singh, 2022). This phenomenon is called “few-shot learning,” which
most common applications of ChatGPT within educational denotes the ability of these models to adapt quickly and efficiently to
environments. new tasks. This proficiency is derived from the extensive general
⋅ To the best of our knowledge, our study is the first large-scale knowledge they acquire during the initial training phase, which can be
qualitative investigation based on actual user feedback that ex­ applied to novel tasks (Beltagy et al., 2022). This transfer learning
plores the viewpoints of early adopters of ChatGPT in the educa­ phenomenon has been witnessed in many contexts, including the crea­
tional context. By illuminating the diverse array of perspectives held tion of sophisticated chatbots and language translation models (Kul­
by early adopters in the education sector, we provide a nuanced karni, Shivananda, & Kulkarni, 2021; Omran, Sharef, Grosan, & Li,
understanding of the positive and negative perceptions surrounding 2023).
ChatGPT.
⋅ The findings from our investigation reveal that ChatGPT is not 2.2. GPT family models
designed to supplant the roles of teachers; instead, it is supposed to
function as an auxiliary tool that assists them in elevating the quality The GPT family of models, Generative Pre-trained Transformers,
of education. The support from educators is crucial for contextual­ introduced by OpenAI, has experienced significant advancements since
izing and authenticating the content produced by ChatGPT. the initial release of GPT-1 in 2018 (Radford, Narasimhan, Salimans, &
⋅ Finally, the key findings from our study can guide the design of Sutskever, 2018; Radford et al., 2019; Brown et al., 2020; Bubeck et al.,
future studies. These could include surveys, interviews, and 2023). These powerful LLMs have demonstrated proficiency in a wide
controlled experiments that aim to better understand the diverse and range of NLP tasks and have grown steadily in size and complexity (Bang
comprehensive views and experiences of different stakeholders when et al., 2023; Bubeck et al., 2023).
incorporating ChatGPT in education. The pioneering model in the GPT series, GPT-1 (Radford, Nar­
asimhan, Salimans, & Sutskever, 2018), was built on the transformer
2. Background and related work architecture and boasted 117 million parameters, allowing it to process
vast amounts of textual information. This inaugural model established a
2.1. Language models solid groundwork for subsequent advancements in the GPT series by
laying the foundation for the exploration and enhancement of
The field of natural language processing, also known as computa­ transformer-based language models. Despite representing a noteworthy
tional linguistics, has made significant progress in recent years, espe­ milestone in natural language processing, GPT-1 exhibited certain
cially with the emergence of large neural network models trained on shortcomings. For instance, the model tended to generate redundant
massive datasets (Zhong, Qian, & Zhang, 2021; Liu et al., 2023). Unlike text, particularly in response to prompts that were beyond the purview
traditional rule-based systems, these large data-driven models achieve of its training data. Moreover, GPT-1 struggled to reason in multiple
greater precision and efficiency in various NLP tasks (Esmaeilzadeh & exchanges within a dialogue and was unable to maintain long-term
Taghva, 2021; Muller, Chilton, Kantosalo, Martin, & Walsh, 2022), such dependencies within the textual content. Furthermore, while the
as language translation (Yang, Wang, & Chu, 2020), sentiment analysis model exhibited cohesion and fluency in short text sequences, these
(Zhang, Wang, & Liu, 2018), named entity recognition (Li, Shang, & qualities diminished in longer passages.
Shao, 2020), and text summarization (Elsaid, Mohammed, Ibrahim, & In February 2019, OpenAI debuted GPT-2 (Radford et al., 2019), a
Sakre, 2022), among others. language model that represented a notable advancement over its pre­
Deep learning has become the dominant approach for developing decessor, boasting 1.5 billion parameters. This model served as a clear
NLP models (Lauriola, Lavelli, & Aiolli, 2022). Among the most popular demonstration of the potential for LLMs to consistently generate text
NLP architectures today are recurrent neural networks (RNNs) (Baktha that is both cogent and contextually appropriate. However, it also raised
& Tripathy, 2017), which include long short-term memory networks concerns about the possibility of its malicious misuse (e.g., fake news
(LSTMs) and gated recurrence units (GRUs) (Baktha & Tripathy, 2017; generation) (Hao, 2019). Despite its numerous merits, GPT-2 was not
Yao & Guan, 2018), as well as state-of-the-art transformer models (Wolf without limitations, as it struggled with tasks that required more com­
et al., 2020). In particular, the transformer model introduced by plex reasoning and contextual understanding.

3
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

GPT-3 (Brown et al., 2020), launched in June 2020, established a and risks. Their research flags several domains where ChatGPT could
new standard in the field with an unparalleled 175 billion parameters. pose challenges, including ethical and social concerns, technical limi­
This model demonstrated an impressive ability to generate human-like tations, humanoversight, and issues of accessibility and the digital
text, enticing researchers and industries to explore its applications in divide.
various NLP tasks (Abdullah, Madain, & Jararweh, 2022; Dale, 2020). Adding to the discourse, Murgia et al. explored the use of ChatGPT in
The GPT-3.5 version further refined the capabilities of the model, which education, with a particular emphasis on adapting online inquiries to
was later used for the development of ChatGPT. meet different literacy levels (Murgia, Pera, Landoni, & Huibers, 2023).
The most recent addition to the GPT series, GPT-4 (OpenAI, 2023), Their findings indicate that while ChatGPT can adjust its responses to
was unveiled on March 14, 2023, and has received widespread attention match a 4th-grade level, there remains room for improvement to achieve
for its exceptional performance and multimodal capabilities. Estimated optimal readability.
to have trillions of parameters, GPT-4 can efficiently process both tex­ Moreover, in self-guided learning environments where ChatGPT
tual and visual inputs, expanding its potential applications and pushing served as a digital mentor, it demonstrated potential benefits for adult
the limits of AI-generated content even further. Although currently learners in defining educational objectives, identifying resources, and
accessible in a restricted capacity, GPT-4 marks a significant advance­ tracking progress (Lin, 2023). However, scholars raised concerns about
ment in the field of large language models. potential over-dependence on ChatGPT and the ambiguity of AI usage
The evolution of the GPT family of models has had a profound impact policies in different educational contexts (Lin, 2023).
on deep learning and NLP research (Zhang et al., 2023). With each new In a societal analysis, Kristjan and Siibak drew attention to the ten­
iteration, these models have demonstrated increasingly sophisticated dency of tech companies to release tools like ChatGPT without fully
language generation capabilities, opening up new research avenues and considering their social implications (Kikerpill & Siibak, 2023). They
applications in various domains (Radford, Narasimhan, Salimans, & noted mixed reactions towards ChatGPT’s introduction–some called for
Sutskever, 2018; Radford et al., 2019; Brown et al., 2020; Bubeck et al., a ban in education, while others perceived it as a “technological fix”
2023; Zhang et al., 2023). demanding a re-evaluation of educators’ roles and teaching philoso­
The GPT family is expected to continue to evolve with the release of phies. They further warned against overlooking the commercial interests
GPT-5, anticipated in early 2024. This development is likely to bring of OpenAI, considering its for-profit status (Kikerpill & Siibak, 2023).
about advances in language understanding, multimodal processing, and In conclusion, while AI and tools like ChatGPT appear to hold sig­
task-specific fine-tuning. These improvements are expected to nificant potential for enhancing learning experiences, concerns about
contribute significantly to the ongoing progress in NLP and AI, driving accuracy, potential over-reliance, and unclarified AI usage policies
transformative changes in research and industry contexts. However, as persist. The current literature lacks a holistic exploration of AI-
the capabilities of the GPT family grow, concerns about potential misuse empowered environments, particularly from the perspectives of stu­
and ethical implications (Ray, 2023; Zhang et al., 2023) are also ex­ dents, teachers, parents, and researchers. Therefore, further research is
pected to increase, especially in the context of education. Balancing the essential to fully comprehend human-AI interaction in educational set­
benefits and risks associated with these powerful LLMs will be para­ tings. Our objective is to provide a comprehensive understanding of the
mount to ensure their responsible deployment and adoption. topic by taking into account the perspectives of a wider spectrum of
stakeholders involved.
2.3. AI in education
3. Research method
AI’s expansive influence across various sectors has had profound
impacts on organizations, societies, and individuals (Dwivedi et al., The purpose of this study is to explore early adopters’ experiences
2023). Dwivedi et al. delved into these implications, examining the and thoughts about integrating ChatGPT into education by analyzing
opportunities and challenges arising from generative AI technologies their social media content on various platforms, i.e., Twitter (recently X),
such as ChatGPT in education, business, and society (Dwivedi et al., Reddit, YouTube, and LinkedIn. Using social media content analysis, we
2023). They underscored AI’s potential to enhance “productivity” across can extract valuable themes and helpful insights from social media data
diverse sectors, while also drawing attention to issues such as privacy on a larger scale (Blandford et al., 2016), ultimately leading to the
threats and misinformation. identification of the primary applications (RQ1) and positive and
Building on this, Calisto et al. investigated the acceptance and negative perceptions of early adopters’ use of ChatGPT in education
application of AI systems, with a particular focus on healthcare settings. (RQ2).
They noted that health-focused AI tools like BreastScreening-AI were The remainder of this section provides detailed information on our
favorably received by target users, contributing to reduced diagnosis data collection process, data description, analysis pipeline, and research
times and fewer errors (Calisto et al., 2022a, 2023). However, they ethics to provide a clear and comprehensive understanding of our
emphasized that trust and acceptance from end-users are pivotal for the qualitative study.
successful implementation of AI techniques (Calisto, Nunes, & Nasci­ Given the advantages and challenges mentioned above of using
mento, 2022). ChatGPT-like chatbots in education, this study sought to identify the
In the realm of education, AI’s role has been predominantly about broader perspectives of early adopters (i.e., researchers, educators,
creating personalized, adaptive learning systems to enhance student learners, parents, and general users (unspecified roles)), about the educa­
experiences (Holmes & Tuomi, 2022). The shift in attention has gone tional applications of ChatGPT to unravel its current deficiencies and
from replacing to empowering teachers, emphasizing their crucial roles strengths and discuss the potentials of future use.
when deploying AI technologies in education (Khosravi et al., 2022).
This is further supported by Shoufan’s research on ChatGPT’s usefulness 3.1. Data collection
for programming tasks, which highlights that AI cannot fully supplant
human intelligence, as evidenced by the unavoidable errors in AI re­ For this study, we leveraged the application programming interfaces
sponses (Shoufan, 2023). (APIs) of Twitter (recently X) (Twitter), Reddit (Reddit), YouTube
A cross-cultural study by Luo et al. provided insights from experts in (Google), and LinkedIn (LinkedIn) to gather data on the use of ChatGPT
China and the United States on the implications of ChatGPT for early in education. Our data collection on LinkedIn was further enhanced by
education (Luo et al., 2023). They concluded that whether ChatGPT’s using the advanced (and legal) data scraping functions available on
impact is beneficial or harmful depends on its implementation and the Helium (v.3) (see Hadi Mogavi et al., 2022).
balance stakeholders strike between its potential benefits, challenges, With over 400 million monthly active users (Ruby, 2023a), Twitter

4
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

provides an ideal platform for screening global users’ opinions on Using this sampling technique, we were able to obtain a diverse but
ChatGPT and its educational applications. Reddit, with more than 50 manageable sample of social media posts. It is essential to note that
million daily active users (Rubyb), provides rich data through its various choosing posts with a high number of views and likes is justifiable
subreddits, including those discussing the use of ChatGPT in education. because they are more likely to contain content that resonates with a
YouTube, with more than 2.5 billion monthly active users (Rubyc), is the larger audience. This is crucial to understanding the dominant attitudes
world’s largest video-sharing platform and offers valuable long-form and conversations on these social media platforms (Tahaei et al., 2020).
audio and visual data, such as seminars, lectures, and crash courses Exclusion Criteria. To ensure the precision and authenticity of our
discussing the use of ChatGPT in education. Finally, LinkedIn, with more analysis, we limited our focus to the experiences of early adopters with
than 300 million monthly active users (Ruby, 2023b), is the world’s real and practical knowledge of ChatGPT in educational settings, which
largest professional networking platform and provides extensive data on were manually verified by three co-authors through user-generated
the professional use of ChatGPT in education. We selected these plat­ content. According to the established literature, we intentionally
forms because of their large user bases, rich data, accommodating terms excluded user comments, external links, and image-based content from
of use, and potential to provide valuable insights into our research our analysis to avoid an exponential increase in data volume. Our
questions about the use of ChatGPT in education (Blandford et al., 2016; analysis solely incorporated comments from the Reddit platform
Sinha et al., 2021). because of its discussion-centric style of communication (see Haq et al.,
Inclusion Criteria. To ensure the relevance of the data collected, we 2022). Examining comments and external links would have greatly
used specific inclusion criteria. The content had to include the keywords complicated our efforts to authenticate and analyze user experiences, as
“ChatGPT” AND (“Education” OR “Learn” OR “Teach” OR “Instruct” OR these sources frequently lack nuanced details of their parent posts.
“Study” OR “School” OR “University”). We chose these keywords based Although we recognize that some of the excluded content may have
on their prevalence of use in learning communities, as determined by provided valuable insights, we considered this exclusion necessary to
Ubersuggest7 (similar to Hadi Mogavi et al., 2021). Our analysis was maintain a more rigorous and systematic approach to data collection
conducted with the objective of minimizing the likelihood of over­ and analysis.
looking crucial information—to the best of our ability—through the use Stopping the Data Collection. As our data collection proceeded in
of an array of keywords. It should be noted that all of the data used in our parallel with our data analysis process, we stopped collecting data once
research were analyzed in text format, but the scope of the data we achieved thematic saturation, which is the point at which no new
collection was restricted to content expressed in English. YouTube data information could be extracted from our data. The decision for thematic
were transcribed into textual format relying on the transcript feature saturation was unanimously agreed upon by all data coders involved in
that is accessible on the platform. In instances where a transcript was not the research, comprising three co-authors. To prevent or reduce the
provided by the content producer, we used the autogenerated transcript chance of premature termination of data collection, we intentionally
functionality offered by YouTube. analyzed a minimum of 1000 pieces of content from each social media
Data sampling. Because studying all social media data is imprac­ platform, although thematic saturation occurred much earlier in most
tical, we followed the advice of existing literature and collected random cases (i.e., on average = 825.69, SD = 145.66). In addition, we used 500
samples from highly viewed and liked social media posts, specifically pieces of content to confirm the accuracy of our thematic saturation as a
those in the top 25% (upper quartiles), on Twitter, Reddit, YouTube, and final measurement.
LinkedIn (refer to Hadi Mogavi et al., 2022; Tahaei, Vaniea, & Saphra,
2020). To accomplish this, we employed a stratified random sampling 3.2. Data description
method, without personalization (see Papadamou et al., 2021), similar
to the method used in a Computer Supported Cooperative Work (CSCW) The data analyzed in this work cover the time period between
study conducted by Fogliato et al. (Fogliato & Chouldechova, 2021). Our December 1st, 2022, and April 22nd, 2023, encompassing the five
decision to use stratified random sampling was motivated by a desire to months following the official public release of ChatGPT. Our study ex­
secure an impartial and equitable representation of early adopters’ amines 6000 text samples totaling roughly 150,000 words to determine
viewpoints, encompassing informed researchers, educators, learners, how people commonly used ChatGPT in various learning contexts.
parents, and general users. We established this by searching for the key­ Compared to similar prominent HCI studies, the size of our dataset
words presented in Table 1 within user profiles and manually annotating seems reasonable to provide a broad view of early users’ opinions (e.g.,
the data. The alternative keywords presented in Table 1 were recom­ see Hadi Mogavi et al., 2021; Pater et al., 2019; Tahaei et al., 2020).
mended using a built-in function designed for similar tasks on the More specifically, our dataset includes 1500 randomly selected posts
popular qualitative research application called Atlas. ti (Atlas, 2023). from each of the four social media platforms (see Section 3.1), with
Prior to stratified sampling, the majority of data only reflected the equal numbers from five user groups in education: researchers (300),
perspectives of researchers, educators, and general users. educators (300), students (300), parents (300), and general users (300).
We used 1000 posts per platform for the thematic analysis, reserving the
Table 1 remaining 500 (100 from each user group) to confirm the final satura­
Alternative keywords for early adopters (in alphabetical order). tion of the themes. Fig. 2 displays the distribution of user discussions
concerning the utility of ChatGPT across diverse educational settings.
Early Adopter Roles Alternative Keywords
According to our analysis, most posts (55.27%) came from users in
Researcher Academic, Analyst, Expert, Investigator, Professional,
North America (34.23%) and Northern Europe (21.04%). The rest came
Scholar, Scientist
Educator Coach, Instructor, Mentor, Professor, Teacher, Trainer,
from: Oceania (4.10%), Asia (2.88%), Africa (1.17%), and elsewhere
Tutor (0.98%). It should be noted that for 35.60% of the data, the user’s
Learner Apprentice, Student, Trainee location is either not provided or disallowed for collection by the social
Parent Father, Guardian, Mother media platform.
General User (unspecified N/A
The use of ChatGPT in education platforms is a more frequently
role)
discussed topic in wealthier, more developed northern countries. This
observation, however, could also be attributed to regional language
differences as well as the dominance of other local social media plat­
forms in other areas. For example, WeChat, Sina Weibo, and Zhihu are
more popular social media platforms in China (Ao et al., 2021; Chen,
7
https://neilpatel.com/ubersuggest/ Wang, Xu, & Huang, 2021; Gong et al., 2021). Furthermore, it should be

5
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

analyzed the data and generated an initial set of codes. The coders then
met remotely via video conference using the Zoom application to review
the codes identified individually, discuss any disagreements in coding,
and arrive at a consensus on the initial set of codes. From this initial
meeting, we compiled a codebook containing a summary of the codes
and examples of data extracts (quotes) for each code.
We went through the independent coding process, followed by group
discussion and consensus, a total of five times over two months. With
each iteration, we refined the codes by collapsing some codes and
separating others into distinct concepts. We also identified emerging
themes throughout the codes. After the fifth round of coding and dis­
cussion, we arrived at a final codebook that contained the most impor­
tant information in the data and answered our research questions.
As a final step in the thematic analysis process, we used the Fleiss’
Kappa measure to estimate and report the level of agreement among all
coders. We calculated the Kappa value using the reliability analysis tool
in the IBM SPSS Statistics10 application. Our study produced a Kappa
score of κ = 0.87, indicating almost unanimous agreement among our
Fig. 2. Distribution of user discussions concerning the utility of ChatGPT across team members on the final themes identified (Hadi Mogavi et al., 2022).
diverse educational settings. ChatGPT is primarily discussed in the context of
higher education, K-12 education, and practical skills learning.
3.4. Research ethics
noted that as of April 25, 2023, ChatGPT is not accessible or has stopped
providing services in countries such as China, Cuba, Iran, Italy, Russia, In this study, we adhered to established ethical guidelines for col­
Syria, Uzbekistan, and Venezuela (Conroy, 2023). lecting and analyzing data from social media platforms (GDPR, 2018;
AoIR, 2019; Hadi Mogavi et al., 2022). Before data collection, we ob­
3.3. Data analysis tained ethical approval from our institution’s research ethics board to
conduct this study. Furthermore, we obtained proper Application Pro­
In this study, we used a qualitative research approach, specifically gramming Interface (API) keys from the platforms to legally access their
the inductive coding method, to analyze the collected social media data. data without overburdening their systems. We meticulously followed
We decided to conduct a qualitative study because this type of research each platform’s terms of service to ensure compliance with platform
is best suited to capture the complex and nuanced aspects of users’ re­ policies and user privacy. Furthermore, we only collected publicly
actions and responses to new technologies, trends, and phenomena (see available data and anonymized them by removing any personally
Blandford, Furniss, & Makri, 2016; Hadi Mogavi et al., 2022). The identifying information. Furthermore, it should be noted that all coders
inductive method is particularly suitable for our investigation because of in this study have received certified training in human subjects protec­
the apparent lack of prior empirical research on user experience related tion regulations, policies, and ethical issues.
to the use of ChatGPT in education. Furthermore, the inductive coding
approach enables us to develop themes from the data without precon­ 4. Findings
ceived notions or prior assumptions (Blandford, Furniss, & Makri, 2016;
Hadi Mogavi et al., 2022). The study’s findings are divided into two sections based on the
We used multiple digital tools to facilitate our coding process. research questions posed: (1) uncovering ChatGPT’s most frequent ap­
Microsoft Excel helped organize and manage our data. Atlas. ti, a plications in education, and (2) exploring different perspectives on
qualitative data analysis software, allowed for collaborative human-AI ChatGPT’s utility in education. The main themes within each section are
sentiment analysis,8 keyword searching, data coding, and report presented as bullet points. Quotes from the content analysis are indi­
writing. Jamboard 9, a virtual whiteboard, allowed brainstorming and cated in italics, and the sources of different quotes and the positions
visualization of emerging themes. The combination of these tools (roles) of each commenter are reported while ensuring anonymity.
enabled a thorough and rigorous analysis of our data. Fig. 3 shows the Moreover, the percentage values indicate the frequency with which a
snapshots of these tools used in our research. topic appears during the content analysis.
Analysis Pipeline. To identify the overarching themes in our data,
we employed Braun and Clarke’s (Braun and Clarke, 2006) six-phase
framework (similar to Jeon, 2023). This involved becoming familiar 4.1. RQ1: exploring the multifaceted uses of ChatGPT in education
with the data, creating initial codes, searching for themes, reviewing
them, consolidating titles and definitions, and finally reporting the Our research reveals that ChatGPT is most frequently used in three
themes. To obtain a more comprehensive understanding of people’s distinct educational settings: higher education (24.18%), K-12 educa­
opinions and minimize the chances of unintentional biases as much as tion (22.09%), and skills training (15.28%). Other areas where ChatGPT
possible, data from all four social media platforms were analyzed has been used include professional development (9.61%), entrepre­
concurrently (similar to Hadi Mogavi et al., 2021). neurship education (7.18%), lifelong learning (4.64%), and popular
Three authors (henceforth: coders) conducted the thematic analysis science education (3.25%). Furthermore, 13.77% of the instances are
through an iterative coding procedure collaboratively. The third coder related to other learning domains that are not explicitly categorized. To
was brought in to help resolve any potential disagreements between the visually represent the widespread use of ChatGPT in education, Fig. 4
first two coders. displays the word clouds from social media-generated content, show­
In the initial phase of coding, each of the three coders independently casing the top 50 applications of ChatGPT within each learning domain.
In the following sections, we explore each category in more detail.

8
The default categories are positive, negative, and neutral
9 10
https://jamboard.google.com/ https://www.ibm.com/products/spss-statistics.

6
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

Fig. 3. Snapshots of tools used during the coding process.

Fig. 4. The word clouds in this figure depict the top fifty applications of ChatGPT within various learning domains, sourced from social media platforms.

4.1.1. Higher education (24.18%) researchers have used ChatGPT to write and proofread research pa­
In the context of higher education, which refers to tertiary education pers, prepare lecture notes, create class presentation slides, and
following secondary education, typically in a college or university compose literature reviews. One professor shared her experience on
setting, ChatGPT has shown significant potential in helping researchers, YouTube, stating:
educators, and students.
“Simply enter the article or book chapter that you want to present in your
• Content Creation and Editing. The most prevalent use of ChatGPT in next meeting or class and then presto! You have a nice outline for your slides,
higher education is content creation and editing, accounting for and the content can be edited however and as how many times as you like!”
78.11% of the reported applications. Educators, students, and

7
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

• Content Summarization. Furthermore, ChatGPT has proven helpful “I’ve been incorporating ChatGPT into my creative writing lessons to
in summarizing academic papers, as shared by a Ph.D. student on create personalized prompts for each student based on their interests and skill
LinkedIn: level. It has not only helped them develop a deeper understanding of story
structure and character development but also motivated them to push their
“ChatGPT is awesome for getting a quick rundown of authors’ works. I’m creative boundaries …. The improvement in their writing and their increased
finding it super helpful lately, and it could be a great way to sift through enthusiasm for the subject are truly remarkable.”
articles for a lit review and figure out what’s [relevant] and worth reading Furthermore, ChatGPT can be integrated into educational online
and what’s not.” games (or game environments), promoting active learning and knowl­
edge retention. A parent shared their experience on YouTube:
• Collaborative Data Analysis. The integration of ChatGPT in higher “ChatGPT has been a great addition to the educational games [on
education has not only streamlined various processes but also mathplayground] my son plays online. It acts as a smart and interactive
inspired new methods of research and data analysis. A distinguished learning assistant, guiding him through challenging math puzzles and
professor of social science and humanities has proclaimed on their providing real-time feedback on his problem-solving skills …. The way
YouTube channel, “ChatGPT is the beginning of a new era of big data ChatGPT adapts to his learning pace and offers customized hints has made
exploration.” ChatGPT can serve as an AI-based “research assistant for learning more enjoyable and effective for him. He’s now more motivated than
text-based data analysis, annotation, and sentiment analysis.” A helpful ever to continue [exploring] and mastering new subjects.”
prompt this professor employs to improve their thematic analysis is
asking ChatGPT, “What are other angles this [issue, topic, or quote] • Supporting Home Learning. ChatGPT has proven to be a valuable
could be analyzed from?” asset for supporting home learning by providing assistance with as­
• Student Evaluation. Teaching Assistants have found value in using signments and helping parents teach complex concepts. By using
ChatGPT for student evaluation. One TA shared on Twitter: ChatGPT, parents can access a wealth of information and resources,
enabling them to better support their children’s educational journey.
“ChatGPT saved me a lot of time and effort while grading 80 C++ A parent shared in a tweet: “ChatGPT has made a huge difference in our
homework submissions! I requested ChatGPT’s assistance in creating a home learning experience. I feel more confident helping my daughter with
grading [rubric], and it was fantastic. I was able to review all the submissions, her math homework because it offers step-by-step guidance and clear
test them more [effectively], and even detect some instances of plagiarism.” explanations. It’s like having a personal tutor at our fingertips!”
• Supporting Accessibility for All Students. ChatGPT has the potential to
• Peer-Review Processes. Finally, the peer-review process is another empower students with disabilities and special needs by providing
area where ChatGPT has made an impact. A professor of HCI shared them with accessible learning resources. For instance, ChatGPT can
his experience on LinkedIn: help students with ADHD (Attention-Deficit/Hyperactivity Disorder)
by offering concise explanations and engaging content tailored to
“Time is money … [So] these days, before agreeing to review [a paper], I their learning style. A special education teacher mentioned on
use ChatGPT to first get an overview of [the work] and assess its overall LinkedIn:
quality. A low ChatGPT score [ChatGPT-based evaluation] will automati­
cally disqualify a paper from my review [list]. ChatGPT does a great job of “ChatGPT has been a game-changer in my classroom, especially for my
summing up the key takeaways and contributions of the research …. I highly students with ADHD. By using ChatGPT, I’ve been able to break down
recommend it to busy colleagues and peer reviewers!” complex topics into shorter, more digestible segments that hold their attention.
Early Adopter Majors. According to our findings, the three main Additionally, the interactive nature of the platform keeps them engaged by
subjects in which ChatGPT has been most frequently utilized by early providing instant feedback and allowing them to work at their own pace. The
adopters in higher education are Social Sciences (15.9%), Business & result is a more focused learning experience, leading to better understanding
Management (11.73%), and STEM Education (Science, Technology, En­ and retention of the material.”
gineering, and Mathematics) (11.63%). Fig. 5 provides a more An educator who works with students with learning disabilities
comprehensive overview of the most popular fields of study in which shared their experience in a Reddit post: “Using ChatGPT, I’ve been able to
ChatGPT is reportedly used the most. develop personalized resources that cater to my students’ unique learning
needs. For example, I can create visual aids and simplified explanations that
4.1.2. K-12 education (22.09%) help them grasp complex concepts more easily. This has made a significant
In K-12 education, which refers to primary and secondary education, difference in their comprehension and overall progress …. By leveraging
ChatGPT has been used to enhance teaching and learning experiences ChatGPT’s capabilities, educators can create inclusive learning environments
for educators, parents, and students. The following are some of the key that cater to the diverse needs of all students.”
applications of ChatGPT in this context: Early Adopter Study Subject. Based on the results of our investi­
gation (refer to Fig. 5), early adopters predominantly employed
• Learning Language Skills. ChatGPT has been instrumental in helping ChatGPT within the K-12 educational domain, concentrating on three
students improve their reading, writing, grammar, sentence- primary subjects: foreign languages (20.22%), programming (12.43%),
building, and conversational language skills. By engaging with and mathematics (11.03%).
ChatGPT, learners can practice their language skills in a low-pressure
environment, enabling them to gain confidence and fluency. An 4.1.3. Practical skills learning (15.28%)
English teacher shared on Twitter: “I’ve been using ChatGPT to help my In the realm of practical skills learning such as “Do It Yourself (DIY)”
students with their vocabulary and grammar. They love the interactive communities (Lin and Huang, 2010), ChatGPT has been utilized to
aspect and it has made a noticeable difference in their writing skills. support people in acquiring and refining hands-on abilities in various
Highly recommended!” fields. Key applications of ChatGPT in this context include:
• Personalized Learning Experiences. ChatGPT can be used to create
tailored learning experiences that cater to individual students’ needs • Problem-solving Assistance. ChatGPT has been influential in assisting
and interests. For instance, teachers can use ChatGPT to construct learners with problem-solving tasks, such as debugging code or
collaborative story-building exercises that help students learn crea­ troubleshooting a malfunctioning device. By providing step-by-step
tive writing skills in a fun and engaging manner. A middle school guidance and engaging in question-and-answer interactions,
teacher shared their experience on LinkedIn: ChatGPT enables users to develop a deeper understanding of the

8
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

Fig. 5. Distribution of ChatGPT usage across different disciplines, study subjects, and learning areas.

issues they face. For instance, one user mentioned on Reddit: “I was 4.1.4. Professional development (9.61%)
struggling with a coding issue for hours, and ChatGPT helped me find the In the context of professional development education, ChatGPT has
bug in no time. The [tool] provided clear instructions and explanations, been employed to enhance a variety of skills and knowledge across
making it easy for me to fix the problem and learn from the experience. multiple sectors. From improving communication abilities to facilitating
• Personal Skill Development. ChatGPT has proven valuable in helping workshop organization, ChatGPT has demonstrated its value in
users develop a wide range of practical skills by providing access to numerous professional settings.
guidance on various projects, techniques, and tools. Examples
include website development, cooking, learning a new language, and • Communication Skills. ChatGPT has so far played a noteworthy role
acquiring information about health and personal care. A freelance in empowering professionals to refine their written and verbal
web developer shared on Twitter: “ChatGPT has been an amazing communication skills, thereby amplifying their effectiveness in their
resource for me as I’ve been learning web development. It offers tips, best respective careers. By providing tailored suggestions and adjust­
practices, and even code snippets to help me polish my skills and create ments, ChatGPT guides individuals in the construction of well-
more professional websites.” organized emails, reports, proposals, and presentations. A project
• Supporting Children’s Skill Development. Parents and teachers can use manager shared their experience on LinkedIn:
ChatGPT to find age-appropriate projects and resources that help
their children develop practical skills in various areas, such as pro­ “ChatGPT has changed how I handle communication, giving me confi­
gramming, storytelling, cooking, or crafting. A parent shared on a dence in my writing and presenting skills. With its smart tips and on-the-spot
Reddit comment: “Thanks to ChatGPT, I’ve found so many awesome feedback, I can write content that’s not just clear and to-the-point, but also
projects for my kids during their school breaks. It gives ideas that match more engaging and convincing …. Thanks to [ChatGPT], my communication
their interests and even explains how to tackle tricky parts step by step. My game has leveled up, and it’s really made a difference in how I connect with
kids have built robots, tried coding, and let their creativity run wild, all people at work.”
while having a blast. I love how ChatGPT makes learning fun and By offering natural language explanations of trends, patterns, and
accessible and helps my family and kids grow their skills and confidence!” correlations, ChatGPT simplifies the process of interpreting and
• Career Exploration. ChatGPT can help users explore different career communicating data-driven findings. A data researcher working at
paths that involve practical skills, providing information about job Microsoft mentioned on LinkedIn:
requirements, necessary qualifications, and potential opportunities. “[ChatGPT] not only helps me identify key insights from large datasets
Users can ask tailored questions like, “What does it take to become an but also communicates those findings in a clear and concise manner that is
AI specialist?” or “How can I excel in a web development career?” A easily understood by my colleagues and clients. It’s like having an AI-powered
college student shared on LinkedIn: “ChatGPT has been an enlightening data storytelling partner on my team!”
guide as I navigate the maze of career paths connected to my major. By
shedding light on the practical skills and qualifications unique to each • Digital Literacy. Equipping individuals with digital literacy skills is
field, I’ve gained a clearer understanding of my options, …empowering me critical in today’s technology-driven world. Reportedly, ChatGPT has
to make well-informed decisions about my time. ChatGPT reveals insights been employed by some industry sectors to facilitate learning in
that often remain hidden beyond the pages of academic textbooks, making various areas of digital literacy, such as software usage, data analysis
it an invaluable ally in the quest for career exploration and skill and management, cybersecurity, and online communication
development.” etiquette for employees. A digital marketing specialist shared on
LinkedIn: “ChatGPT has proven to be a valuable tool for my professional
Early Adopter Learning Areas. As shown in Fig. 5, our analysis development. It enables me to rapidly understand new concepts, while
reveals that within the realm of practical skills learning, early adopters simultaneously enhancing my decision-making abilities for my clients’
have predominantly harnessed the power of ChatGPT to enrich their benefit. It’s akin to having a personal digital literacy tutor, accessible 24/
knowledge and expertise in the areas of programming and coding 7 for tailored guidance and support.”
(17.61%), health and self-care (14.28%), and web design and development • Problem-solving. ChatGPT has been used to boost problem-solving
(14.03%). skills in various professional settings. By asking questions and
exploring different solutions, users can work with ChatGPT to un­
derstand complex problems and come up with creative ideas. A

9
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

business analyst mentioned in a Reddit post: “Lately, when I’m stuck reducing cart abandonment rates, which we hadn’t considered before. While
with a tough problem at work, I turn to ChatGPT for help. It gets me to some information may occasionally be less relevant, ChatGPT consistently
think outside the box and see other viewpoints that I might not have provides valuable insights that enhance our benchmarking comparisons,
thought of. The interactive and collaborative vibe of ChatGPT has really enabling us to gain a deeper understanding of our market position and make
improved my problem-solving skills and overall job performance and data-driven strategic decisions with increased confidence.”
happiness.”
• Workshop Organization. Additionally, some users have mentioned • Proposal Writing. ChatGPT has been utilized by entrepreneurs and
that ChatGPT has been a helpful tool in organizing and facilitating their teams to develop convincing grant proposals and pitch decks.
workshops, training sessions, and seminars. By creating agendas, By offering support in conducting literature reviews, proposing en­
activity plans, and resource materials, ChatGPT simplifies the plan­ hancements in structure, and recommending pertinent (though
ning process, letting facilitators concentrate on providing engaging possibly not the most recent) references, ChatGPT assists users in
and effective learning experiences. A corporate trainer shared their crafting compelling content that appeals to potential investors and
experience on YouTube: stakeholders. A grant writer for a social enterprise shared their
experience on YouTube: “It’s awesome to find tools like ChatGPT that
“ChatGPT has been quite useful in organizing workshops for my clients. make grant proposal writing easier. It’s great at finding relevant literature
For instance, while brainstorming activities, it suggested a hands-on group and organizing content, which saves a ton of effort. But its real strength is
exercise to encourage creative thinking, which was well-received by partici­ creating a persuasive language that makes proposals shine. This is key for
pants. It also assisted in creating detailed agendas and training materials, grant writing, as you gotta show why your project deserves funding. With
reducing my preparation time and allowing me to concentrate on delivering ChatGPT, you can be sure your proposals have a better shot at success, …
high-quality content …. The feedback from participants has been predomi­ even when writing in a language other than English.”
nantly positive, and I believe a significant portion of that success can be
attributed to the support I’ve received from [ChatGPT].” The wide-ranging applications of ChatGPT in entrepreneurship ed­
Early Adopter Learning Areas. As depicted in Fig. 5, our findings ucation and apprenticeship programs emphasize its adaptability and
demonstrate that, in the context of professional development, early capacity to empower future entrepreneurs and innovators. This, in turn,
users have mainly utilized ChatGPT to enhance their understanding and cultivates a more vibrant and prosperous entrepreneurial ecosystem.
proficiency in the fields of Information Technology (IT) and Digital Liter­ Early Adopter Learning Areas. As illustrated in Fig. 5, our research
acy (17.08%), Design Thinking11 (7.29%), and Business Writing Skills shows that within entrepreneurship education and apprenticeship pro­
(6.96%). grams, early adopters predominantly use ChatGPT to improve their
knowledge and skills in Design Thinking (12.06%), Critical Thinking
4.1.5. Entrepreneurship education and apprenticeship programs (7.18%) (11.05%), and Business Modeling (9.65%).
In the domain of entrepreneurial skills and expertise, ChatGPT has
emerged as an invaluable information resource for entrepreneurs and 4.1.6. Lifelong learning (4.64%)
startups seeking to innovate, develop, and expand their businesses. In the context of lifelong learning, ChatGPT has emerged as a valu­
Spanning across brainstorming, benchmarking, and proposal writing, able tool that supports self-directed learning by offering a diverse range
ChatGPT has reportedly played a beneficial role in numerous facets of of resources and features customized to individual needs. These features
the entrepreneurial journey. include facilitating microlearning experiences, recommending custom­
ized learning paths, and providing ongoing mentoring support.
• Brainstorming. ChatGPT has demonstrated its usefulness for entre­
preneurs by generating innovative ideas and presenting various • Microlearning. ChatGPT excels at delivering engaging, bite-sized
viewpoints. It aids the ideation process by suggesting enhancements, learning experiences through interactive question-and-answer ses­
pinpointing potential issues, and fostering creativity. A startup sions, concise lessons, and clear explanations of a wide array of
founder shared their experience on YouTube: “With ChatGPT phenomena. This targeted approach enables learners to easily grasp
involved, it’s like having an additional team member who brings a wide and retain information, making it particularly beneficial for those
range of diverse ideas, enriching [our] group’s creative thinking …. pressed for time or looking to deepen their understanding of specific
ChatGPT has been a helpful addition to our brainstorming sessions, of­ topics.
fering valuable insights and [alternative] approaches that led to positive,
profitable outcomes. For example, [it] once suggested an interesting A YouTuber, who identifies himself as a lifelong learner, shared his
marketing strategy targeting a previously neglected customer segment [in enthusiasm for ChatGPT on his channel: “As someone constantly juggling
the user feedback pool], resulting in success …. Although we’ve seen tasks, I’ve found that ChatGPT is exceptional at quickly explaining various
encouraging results, we understand the need for ongoing evaluation and subjects, from Photoshop techniques and video editing tips to makeup tuto­
tracking its performance over time.” rials. The ability to ask questions and receive straightforward, focused an­
• Benchmarking. ChatGPT can help entrepreneurs and students gain swers has made the learning experience with ChatGPT far more enjoyable
insights into industry trends, establish benchmarks, and determine than sifting through search engine results or leafing through magazines.” He
essential performance indicators for comparative analysis. While the concludes, “Although ChatGPT may not always offer the most current in­
information supplied may occasionally be somewhat dated owing to formation or be entirely accurate, it remains an invaluable resource for
ChatGPT’s training data constraints, the AI assistant remains capable broadening one’s general knowledge as well as delving deeper into specific
of delivering valuable insights into prevailing industry standards and areas of interest or questions.”
effective benchmarking practices. A co-founder of a tech startup
shared on LinkedIn: • Learning Path Suggestion. ChatGPT can recommend customized
learning paths based on an individual’s self-declared goals and in­
“Using ChatGPT, we’ve been able to identify crucial benchmarks and terests, such as establishing mastery in a programming language or
performance metrics specific to the e-commerce industry …. [Among its natural language. By providing guidance on the “necessary learning
contributions,] ChatGPT assisted us discover innovative strategies for requirements, resources, and steps,” ChatGPT enhances the learning
experience and helps users achieve their desired outcomes. A soft­
ware developer shared their experience on a tech-oriented Subreddit:
11
As a problem-solving framework.

10
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

“I was eager to learn [a new programming language] but had no idea incredible boon for communicating complicated scientific concepts in our
where to begin. ChatGPT showed me the ropes, pointing out key concepts, online seminars and classes.”
resources, and steps to [mastering] the language in a matter of weeks. By • Storytelling for Education. In addition to simplifying complex con­
following its [tailored] recommendations, I was able to create a personalized cepts, effective popular science education often relies on storytelling
learning plan that met [my needs] and allowed me to [advance] as a pro­ to captivate and inform audiences. ChatGPT’s natural language
fessional at my own pace.” processing capabilities make it well-suited to crafting engaging
narratives that weave together scientific concepts and real-world
• Continuous Mentoring. ChatGPT serves as a valuable resource for applications. A YouTube science educator shared their thoughts on
lifelong learners by offering comprehensive support, including ac­ the platform: ”Incorporating ChatGPT into my video creation process has
cess to information, resources, and constructive feedback. Whether it not only helped me simplify difficult topics but also craft compelling
involves refining a written piece, nurturing creative ideas, managing stories that hold my audience’s attention. The combination of clear ex­
projects, or drafting research findings, ChatGPT effectively functions planations and engaging narratives has significantly improved the quality
as a consistent mentor. Notably, older adults with limited literacy of my content and contributed to an increase in viewership.”
skills have found ChatGPT particularly inspiring, as it enables them • Scientific Visualization. The visual representation of scientific con­
to continue learning without feeling shy about asking basic ques­ cepts poses a significant challenge to scientific content creators.
tions. Moreover, several laborers have credited ChatGPT for moti­ However, it is crucial for enhancing comprehension and engagement
vating them to pursue vocational education, given the platform’s in popular science education. ChatGPT can aid in generating inno­
availability and accessibility as a constant educational companion. vative ideas for graphs, web-based interactive elements, slide de­
This highlights the substantial impact ChatGPT has on diverse signs, and other visualization techniques. Recently, ChatGPT has
learner profiles, providing tailored and timely support to individuals been utilized in conjunction with generative AI tools such as Mid­
from various backgrounds. However, it is widely noted among users journey12 and Stable Diffusion,13 enabling the creation of smart
that ChatGPT’s inability to remember previous contexts poses a prompts, and resulting in more customized image generation. These
challenge that must be addressed in order to provide more effective features effectively help transform complex information into more
and personalized mentoring for lifelong learners. tangible and accessible visual formats. A science writer on LinkedIn
shared their experience, emphasizing the value of ChatGPT in the
A writer seeking to develop their skills with ChatGPT’s assistance realm of popular science education: “Working with ChatGPT has
shared their experience on YouTube: “[ChatGPT] provides constructive allowed me to come up with engaging visualizations for my [research-
criticism and actionable advice for enhancing my work, contributing to my based] articles …. [Just recently,] in one of our collaborative attempts,
growth as a writer. However, its inability to remember previous interactions it came up with a clever analogy for a complex research concept in one of
can be tedious, as I often need to re-establish context. Despite this limitation, I my [pieces,] and I turned it into an eye-catching infographic …. When I
believe ChatGPT remains a valuable asset that has had a positive impact on presented it at a public science [exhibition,] people found it both easy to
my writing journey.” understand and visually appealing.”
In summary, ChatGPT’s applications in lifelong learning demon­
strate its potential to empower individuals throughout their continuous Early Adopter Learning Areas. As illustrated in Fig. 5, our research
growth and development. demonstrates that early adopters in popular science education pre­
Early Adopter Learning Areas. As depicted in Fig. 5, our study dominantly utilize ChatGPT to bolster their knowledge and skills in the
reveals that, in the realm of lifelong (continuous) learning, early fields of Computer Science and Technology (20.62%), Astronomy
adopters chiefly employ ChatGPT to enhance their proficiency and (19.50%), and Biology (14.18%).
expertise in Languages (21.91%), Technology Domain (16.22%), and
Finance (9.06%). 4.1.8. General education (uncategorized) (13.77%)
While many ChatGPT applications can be readily classified into
4.1.7. Popular science education (3.25%) specific educational and learning domains, certain instances of its use
Popular science education plays a crucial role in making complex remain uncategorized due to the lack of (sufficient) context in social
scientific concepts accessible and engaging for general and often non- media content. Since ChatGPT applications such as creative and critical
expert audiences (Xia, Ng, Chen, & Hollan, 2022). It incorporates content writing and editing, text summarization, and providing helpful
various media forms, such as YouTube videos and Medium articles, to learning suggestions have been previously discussed and covered in
effectively communicate scientific ideas, including topics like quantum earlier sections, this subsection will focus on three general topics that
physics, quantum computing, or simply the water cycle. By simplifying emerge more prominently within the context of uncategorized dis­
abstract concepts and promoting scientific literacy, popular science courses on ChatGPT’s educational applications: customized curriculum
education fosters a greater appreciation for the scientific method and its design, student assessment, and fostering collaboration.
contributions to society (Xia et al., 2022). According to our findings,
ChatGPT has become a valuable tool in this domain, assisting content • Customized Curriculum Design. ChatGPT has the potential to help
creators and educators in achieving their educational objectives. educators develop customized and adaptive curricula. By under­
standing the specific learning objectives and desired outcomes in an
• Simplifying Complex Concepts. ChatGPT excels at breaking down educational program (or a course of study), ChatGPT can help con­
intricate scientific ideas into digestible and more comprehensible nect the dots and recommend customized instructional resources and
explanations. By providing analogies, examples, and concise defini­ approaches tailored to meet different learning goals. However, a
tions, it enhances the overall understanding of scientific concepts for human educator would still need to guide the overall curriculum
a broader audience. A science communicator shared their experience design process, ensure the curriculum meets key requirements and
on LinkedIn: “Using ChatGPT, I recently explained gravitational waves standards, and incorporate human judgment in determining the
to a group of public audience members by comparing them to ripples in a optimal learning paths and experiences for students. An instructional
pond. This comparison struck a chord with my audience, piqued their designer conveyed their experience on YouTube, stating: “AI won’t
curiosity, and made the topic appear more accessible and engaging.
ChatGPT’s potential for making complex and abstract ideas easier to
understand while maintaining a high level of accuracy has been an 12
https://www.midjourney.com/home/.
13
https://stablediffusionweb.com/.

11
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

replace instructional designers, but it’ll definitely change our role. With various social media platforms. Our findings suggest that ChatGPT can
systems like ChatGPT, we can now create, or rather draft, curricula that potentially yield both positive and negative outcomes in terms of pro­
are personalized, adaptive, and responsive to various learning objectives ductivity, efficiency, and ethics.
super quickly. Even so, human judgment stays crucial …. [This means] Early adopters of ChatGPT within the educational sector have re­
instructional designers will have an even bigger responsibility as the ported mixed perceptions. Specifically, 30.62% expressed exclusively
planners of human-AI collaboration in education.” positive views, 17.15% exclusively negative views, 37.34% mixed views
• Student Assessment. ChatGPT’s capacity to generate tailored (encompassing both positive and negative aspects), and 14.89% main­
(practice) exams and homework assignments, covering a broad range tained a neutral stance. Table 2 presents a comprehensive illustration of
of information and skills, provides a useful resource for educators, the distribution of individuals’ perspectives concerning the implications
parents, and students themselves to monitor learning progress of ChatGPT on productivity, efficiency, and adherence to ethical prin­
effectively. An educator remarked on YouTube, “Previously, the idea ciples within various educational settings. In the subsequent sections,
of designing personalized tests appeared unattainable; however, that has we explain these diverse perspectives on the integration of ChatGPT into
changed. Utilizing ChatGPT, I’ve created [assessments] that address education in greater detail.
multiple [cognitive] levels and incorporate relevant real-world examples,
which helps my students connect with the material, [fostering] a more 4.2.1. The spectrum of opinions on productivity
authentic learning experience.” In the context of this study, productivity refers to the capacity of
• Facilitating Collaboration. ChatGPT can play a crucial role in ChatGPT to enhance the quantity of education-related and learning
fostering effective communication and collaboration, especially in outcomes within a specific timeframe. The spectrum of opinions on
diverse and multilingual settings. By providing translation services productivity can be divided into four primary groups:
and facilitating the exchange of ideas, it helps create an inclusive • Positive Impacts on Productivity. Our findings reveal that, with the
learning environment where students feel empowered to work exception of K-12 education, entrepreneurship, and general education
together and share their insights. A user shared on Reddit, “This is (uncategorized), individuals generally possess a positive perception of
what worldwide teamwork for learning could look like in the future. With ChatGPT’s utility for improving productivity in the educational domain.
ChatGPT’s ability to understand different languages, obstacles to con­ According to our findings, these favorable views on productivity largely
necting with each other disappear and opportunities to work together stem from ChatGPT’s capability to streamline various educational pro­
arise. Students who struggle to say what’s really on their mind can have cesses, such as content creation, editing, and text summarization.
their ideas translated and spread, giving them a space to be heard and An educator’s post on LinkedIn exemplifies this positive sentiment at
making co-creating new knowledge a reality.” its best: “Ever since using ChatGPT, I’ve created and updated course mate­
rials much faster. It’s made me more productive. For teachers who spend
Although ChatGPT shows promise across a variety of educational much personal time on admin work, this could free them to focus on
fields, it is important to note that in the general education category students.”
(unspecified), the majority of learning areas, specifically more than • Negative Impacts on Productivity. Negative opinions are primarily
85%, lack precise definitions or a concentrated focus, resulting in found in the context of K-12 education. Teachers and parents express
broader discussions on education. Moving forward, we explore RQ2, concern that ChatGPT may exacerbate procrastination, leading students
examining both the positive and negative perspectives of using ChatGPT to become increasingly disengaged from their tasks and less productive.
in education to provide a more complete picture of its potential benefits A teacher elucidated this point on YouTube, stating, “Lazy people may
and challenges. become even lazier and possibly not do anything at all …. The thought that the
task can be completed at any time can lead to doing nothing at all.”
These concerns can be better understood in light of the well-known
4.2. RQ2: promises and Perils of ChatGPT in education theory of Self-Determination (Tyack and Mekler, 2020), which posits
that individuals are more motivated and engaged when they experience
Our analysis of social media discussions reveals that productivity autonomy, competence, and relatedness in their activities. In educa­
(73.19%), efficiency (63.02%), and ethics (48.51%) are the predomi­ tional settings, students who rely heavily on ChatGPT may feel a
nant topics related to the integration of ChatGPT into education. In­ decreased sense of autonomy and competence, as the AI tool assumes
dividuals have expressed diverse opinions on these aspects across

Table 2
Frequency distribution of people’s perceptions regarding ChatGPT’s impact on productivity, efficiency, and ethical values across various educational
settings.

12
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

most of the responsibility for their learning and academic success. This asking, or sometimes you have trouble communicating your thoughts in a way
diminished sense of self-determination could lead to lower motivation the other person can fully understand. ChatGPT has no problems with that.”
and engagement, making students more passive and less productive in • Negative Impacts on Efficiency. Negative opinions regarding
their education. ChatGPT are mainly found in the contexts of higher education and K-12
• Mixed Views on Productivity. In entrepreneurship education and education. Notably, while most early adopters in higher education
general education (uncategorized), opinions tend to be mixed. Users believe that ChatGPT can enhance productivity, they argue that it might
acknowledge both the benefits and drawbacks of using ChatGPT in ed­ not be as effective in terms of learning quality, particularly for advanced,
ucation, suggesting that its impact on productivity depends on certain intellectual, and creative subjects.
factors, most notably the prompt design and the ability to verify or In the K-12 context, concerns about efficiency arise due to ChatGPT’s
organize generated content. Poorly designed prompts and inadequate potential to impede the development of critical thinking and problem-
domain knowledge may result in time-consuming and disorganized solving skills among students. Some educators contend that students
outcomes, which could ultimately hinder productivity. might depend too heavily on AI-generated responses, resulting in a lack
One user on YouTube expressed this sentiment the best by sharing, of deep understanding and an inability to apply knowledge in novel
“ChatGPT can be a great help when it comes to brainstorming ideas or situations. Excessive use of ChatGPT could also diminish interaction
drafting content, potentially speeding up the process. However, you still need between students and teachers, which they deem an essential element of
to invest some time refining the output and making sure it’s accurate …. While the learning process.
it may not be a magic solution for productivity, it can be a valuable tool if used A further concern in both higher education and K-12 settings is the
properly.” possibility of students not verifying the validity of the information
• Neutral Views on Productivity. Neutral perspectives do not explicitly provided by ChatGPT. As AI-generated content may not always be ac­
express a positive or negative opinion on ChatGPT’s impact on pro­ curate, students could inadvertently accept incorrect information as
ductivity in education. Instead, they report potential outcomes and call true, negatively impacting their learning quality.
for further research, experimentation, and observation to determine Wrapping up, a student’s comment on Reddit aptly illustrates this
ChatGPT’s true influence on educational productivity. negative viewpoint: “I bought GPT-4 to help me explain and work through
An example of a neutral view shared by a researcher on Twitter is: problems in my Calculus 1 class, but it hasn’t been able to solve even basic
“As AI writing tools and models like ChatGPT grow more sophisticated, we math equations that involve things like related rates, linearization, etc. It’s
need more data and research to understand their full impact on productivity pretty disappointing to be honest. I heard GPT-4 was connected to Wolfram
in education.” Alpha now and I thought it would be reliable. It still confidently provides very,
very wrong answers for almost every Calculus 1 problem I ask it to solve.”
4.2.2. The spectrum of opinions on efficiency • Mixed Views on Efficiency. Consistent with the mixed opinions on
In the context of this study, efficiency refers to ChatGPT’s ability to productivity, early adopters also exhibit mixed perspectives regarding
enhance the quality of educational processes and outcomes, including the utility of ChatGPT for efficiency in the context of entrepreneurship
improvements in learning, expertise, and skill development. As with education and other general (uncategorized) educational domains.
productivity, the spectrum of opinions on efficiency can be divided into Users believe that ChatGPT can be both a valuable resource for
four primary groups: enhancing learning experiences and a potential distraction, depending
• Positive Impacts on Efficiency. Our findings show that in the on how it is utilized. An apt illustration of such opinions can be found in
educational settings of practical skills learning, professional develop­ a post by an early adopter on LinkedIn: “ChatGPT is super helpful for quick
ment, lifelong learning, and popular science education, most users answers and tackling tough topics, but it’s not foolproof. That’s why it’s
possess positive perceptions toward the utility of ChatGPT for improving crucial to rely on our own thinking too. For effective learning, we need to
the efficiency of learning. merge the AI’s input with our critical thinking and problem-solving skills. This
These positive views are often attributed to ChatGPT’s capacity to approach helps us truly grasp the subject matter and ensures we’re not solely
provide customized feedback, generate relevant examples, and offer dependent on ChatGPT.”
quick and simple answers to a wide variety of questions. These features Aligning with this perspective, a compelling instance from the
can help enhance students’ understanding of complex concepts, facili­ applied educational sphere of entrepreneurship education is shared by
tate skill development, and support educators in delivering tailored an early adopter on YouTube: “As a beginner, ChatGPT has really helped
instruction. me learn a lot about the specific language of entrepreneurship and how to give
The statement of a self-proclaimed lifelong learner on LinkedIn il­ clearer feedback to potential clients …. However, it’s important to note that it
lustrates this positive sentiment at its best: “I’m already noticing my doesn’t quite provide a practical understanding of [the concepts it teaches.]
conversation skills improving. I’ve been using ChatGPT to learn a new lan­ So, I think asking experienced professionals for advice and attending work­
guage by discussing interesting everyday topics in my target language …. I ask shops to fully dive into the business world atmosphere are still necessary and
it to act as an expert on a topic I enjoy, and then it uses all the specific terms, cannot be fully substituted by ChatGPT.”
vocabulary, and slang while responding in the language I’m learning …. It’s • Neutral Views on Efficiency. Similar to neutral perspectives on pro­
really engaging and keeps me [motivated] because I get to learn all the ductivity, neutral views on efficiency do not explicitly express a positive
important words I need to be fluent in the subject!” or negative opinion on ChatGPT’s impact on efficiency in education.
Another early user on Reddit provides a similar, more detailed Instead, they report potential outcomes and call for further research,
experience sharing: “I’ve been working on a coding project and I ask experimentation, and observation to determine ChatGPT’s true influ­
[ChatGPT] all the dumb questions I’m too embarrassed to put online. It’s ence on educational efficiency.
great because it teaches at my own pace and has infinite patience. I can run An example of a neutral view shared by a researcher on Twitter is:
solutions by it, ask if they will work, and if not, why not, and if there are any “The outcome of ChatGPT on student learning hasn’t been conclusively
issues I haven’t foreseen. I use it to brainstorm, I just start typing out ideas and determined. Further research and input are necessary to understand its actual
it gives me input. If I don’t know how to tackle a certain part of the project, I role and to establish an appropriate balance between human-AI interactions
write out the problem and ask it for suggestions and pros and cons. If I don’t in learning experiences.”
understand a concept, I can tell it what I think the concept is and ask if I’ve got
it right. If it explains something to me that I don’t fully understand, I can ask it 4.2.3. The spectrum of opinions on ethics
for examples, or I can rephrase it to the AI and it will see where I went wrong/ In the context of this study, ethics refers to the moral implications
what I missed. It’s a lot better than people for answering those kinds of and concerns associated with the use of ChatGPT in educational pro­
questions because people aren’t always paying attention to what you’re cesses. Similar to productivity and efficiency, the spectrum of opinions

13
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

on ethics can be categorized into four primary groups: • Mixed Views on Ethics. In line with the heterogeneous opinions
• Positive Impacts on Ethics. In line with our previous findings on about productivity and efficiency, the majority of early adopters express
productivity and efficiency, most early ChatGPT adopters in the mixed views about the impact of ChatGPT on promoting or destroying
educational settings of practical skills learning, professional develop­ ethical values in the educational context of entrepreneurship and in
ment, lifelong learning, and popular science education believe that the other (uncategorized) educational settings. A comment from a business
tool will help refine ethics in the education sector over time. educator on YouTube is a good example that illustrates the mixed views
These proponents base their arguments on several key factors. Our on using ChatGPT: “ChatGPT is an exceptional tool for communication, but
findings indicate that some early adopters appreciate ChatGPT’s po­ it shouldn’t replace hard work or be used for dishonest acts like intellectual
tential to democratize access to education and foster equitable learning theft …. This groundbreaking technology can help professionals who struggle
opportunities for students globally. They assert that the AI’s capacity to to express their ideas clearly. By using ChatGPT to generate alternatives or
deliver personalized learning experiences and provide instantaneous expand on concepts, they can better share their knowledge with others ….
feedback can help bridge the divide between students from diverse so­ Sadly, some may misuse ChatGPT for unethical reasons like plagiarism or
cioeconomic backgrounds and those with varying learning needs. taking credit for others’ work. It’s important to remember that the true value
Among these supporters, some specifically emphasize that ChatGPT of ChatGPT is to supportand improve your own efforts, not to replace them.”
could be particularly beneficial for learners with disabilities or those • Neutral Views on Ethics Regarding ethics, neutral views do not
requiring special education. explicitly endorse or criticize ChatGPT’s influence on ethical consider­
Furthermore, these supporters underscore the role of ChatGPT in ations in education, similar to the perspectives on productivity and ef­
cultivating motivation for self-education among individuals of all ages ficiency. Based on our analysis, neutral comments predominantly take
and at any time, thus empowering learners to become more independent the form of questions, as people want to know what is considered
and take charge of their own educational journey in a more confident plagiarism with ChatGPT and what is not, what is ethical and what is
and engaging manner. not, and what the boundaries are. In a tweet, for example, a high school
The optimistic view is probably best reflected in a tweet from an student asks, “Is using AI models like ChatGPT to assist in developing and
early adopter employee, stating: “ChatGPT makes education easy to ac­ enhancing essays and research papers considered an acceptable approach?”
cess, personalized, and open to everyone. It helps people from different
backgrounds and abilities to learn and work together, breaking down the 4.3. Revisiting ethics in the context of AI-driven education
usual barriers.”
• Negative Impacts on Ethics. Our investigation of ChatGPT’s impact As noted earlier, AI’s potential to democratize education is prom­
on ethics primarily highlights concerns in the context of higher educa­ ising. However, according to some critics, it raises the concern of a
tion and K-12 education. The top three areas of ethical concern potential homogenization of thought, as AI models are predominantly
encompass academic integrity, the spread of misinformation and fraud, trained on widely available data. This could lead to a subtle standardi­
and privacy and data security issues. zation of knowledge and thinking patterns under the guise of
One of the most significant ethical issues pertains to ChatGPT’s po­ ‘personalization.’
tential to promote academic dishonesty. Specifically, ChatGPT’s ability The possibility of academic dishonesty and misinformation dissem­
to produce contextually coherent text enables students to easily ination challenges our traditional notions of ‘knowledge’ and ‘learning.’
plagiarize, paraphrase, or distort content in their writing assignments When AI can generate contextually relevant and coherent text, we need
and exams. This compromises academic integrity and undermines the to reconsider what it means to ‘know’ something or ‘learn’ something.
educational value of specific assessment methods. Consequently, in­ This suggests the need to shift the focus of education from mere
structors face immense difficulty in accurately assessing students’ knowledge acquisition to fostering critical thinking, creativity, and
genuine understanding and skills. emotional intelligence—areas where AI still falls short.
Another ethical concern involves ChatGPT’s inadvertent propaga­ Data privacy and security concerns in AI-integrated education go
tion of misinformation and fraud. Although AI models like ChatGPT beyond safeguarding data. They bring to the fore the power dynamics
strive to generate relevant content, they may inadvertently produce inherent in data ownership. Questions arise about who should own the
deceptive or erroneous information due to training data limitations or data generated from students’ learning and who should be responsible
inherent biases. This raises apprehension regarding the dependability of for potential misuse. This necessitates a thorough exploration of data
ChatGPT-generated content and its potential to mislead students, rights and ownership in an AI context.
particularly in the absence of rigorous fact-verification mechanisms. The perspectives offering mixed and neutral views highlight the
Finally, incorporating AI tools like ChatGPT into pedagogical prac­ complex relationship between AI and human agency. The ethical
tice necessitates a strong focus on privacy and data security. AI models conundrum lies in the blurred boundaries between AI as a tool to extend
generally require vast amounts of data to operate effectively. Some early human capabilities and AI as an autonomous entity capable of impacting
adopters express concerns that without well-defined protection policies human lives.
and limited information about ChatGPT’s inner workings, there is a risk Moreover, the biases inherent in AI could reflect our societal preju­
of exposing personal data and records of tool usage to potential dices, effectively holding a mirror up to societal norms (Foroughi et al.,
breaches, unintended applications, or even misuse. 2023). This revelation can serve as a catalyst for critical discussions
In conclusion, a grad school instructor’s personal experience shared about hidden biases in our daily lives and how we can challenge and
on Reddit best exemplifies some of the concerns and challenges sur­ address these biases. This realization underscores the importance of
rounding the use of AI-generated content in academia: “Students had a maintaining a “human-in-the-loop” approach in AI systems (Calisto,
project report milestone to submit and just reading through the submissions I Santiago, Nunes, & Nascimento, 2021). By actively involving humans in
could see ChatGPT written all over it. My students who are international the decision-making processes of AI systems, we can ensure that the
students that struggle to communicate with me in emails all started writing technology is supervised, guided, and corrected by human intelligence
like native speakers in their reports. But that’s not enough proof that they and ethical judgment.
cheated, right? Then the fun begins. All 15 students are citing papers that To sum up this section, the ethical challenges posed by the use of AI
don’t exist. Five to six citations and references and not a single one is a real in education provide an opportunity for societal reflection and trans­
paper which gave me enough proof to fail them on the project and accuse them formation. Our response to these challenges will shape not only the
of using ChatGPT. These students account for 1/3 of my class who are now future of education but also our society’s trajectory in an AI-driven
more or less failing the class half way through the semester. A lot of the world. The ethical considerations of AI use in education merit
remaining 2/3 of the class are already struggling.” ongoing, critical, and nuanced dialogue among all stakeholders.

14
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

5. Recommendations 2022).
To cultivate digital literacy, grasping technology’s multifaceted ca­
This study conducted an in-depth analysis of social media content pabilities and fundamental mechanics is imperative. Users should have a
from four platforms to investigate early adopters’ use and perceptions of holistic view of the underlying principles and operations of data-driven
ChatGPT in education. Our findings suggest that users utilize ChatGPT computational models and algorithms to maintain realistic expectations
for various reasons and perceive it as a valuable educational tool that of tools such as ChatGPT. With this knowledge, users can discern these
can assist students and educators with various time-intensive tasks, such technologies’ potential applications and limitations in educational
as research, problem-solving, customized feedback and language contexts more effectively. For instance, it is noteworthy that although
learning. However, early adopters also expressed concerns about the ChatGPT can produce coherent and contextually relevant text, it may
potential misuse, including cheating, overreliance on the AI tool for encounter difficulties with nuanced or highly specialized subjects and
answers, indolence and superficial learning. technical inferences. As a result, using ChatGPT as the main reference
An essential inquiry arising in this context is the optimal utilization for drafting legal documents is generally inadvisable; however,
of ChatGPT to enhance and facilitate educational objectives. Based on employing ChatGPT to provide an initial outline or suggestions for legal
the collective insights derived from research questions RQ1 and RQ2, we content under the supervision of legal professionals might be acceptable.
put forth the subsequent recommendations to promote the effective Similarly, seeking medical advice from ChatGPT is generally inadvis­
incorporation of ChatGPT in educational settings while concurrently able, but employing ChatGPT to assist patients in articulating their
addressing some of the relevant challenges. Fig. 6 provides an overview thoughts under the guidance of medical professionals might be accept­
of all recommendations in a flowchart. able. Thus, the bottom line is that users must exercise prudence when
assessing the accuracy and pertinence of the generated content.
5.1. Building digital literacy for responsible AI integration in education AI systems may unintentionally perpetuate and reinforce existing
biases in the source material by analyzing vast but limited datasets (Rae
Digital literacy encompasses the competencies and skills necessary et al., 2021). Hence, educators and learners must develop skills to
for individuals to effectively and responsibly engage with complex in­ investigate, identify and address these biases where possible. Applying
formation ecosystems (Tinmaz, Lee, & Fanea-Ivanovici, 2022; Yi-No such knowledge when integrating ChatGPT into learning platforms (for
Kang, Chen, & Chen, 2023). In today’s rapidly evolving educational example, question pool websites, massive open online courses (MOOCs)
landscape, digital literacy is essential for educators and learners to and virtual learning environments) is crucial to foster a fair and inclu­
adeptly utilize technology and adopt the appropriate growth mindset sive learning environment. Some instructors have suggested that critical
(Liang et al., 2021). This is particularly significant when interacting with thinking, media literacy and the comprehension of the ethical implica­
artificial intelligence tools, such as ChatGPT, where digital literacy is tions of AI can serve as effective ways to develop these skills.
pivotal in identifying and addressing the emerging ethical, legal and Digital literacy also encompasses the ethical dimensions of AI usage,
societal implications (Yang, 2022; Tinmaz, Lee, & Fanea-Ivanovici, including data privacy and security, academic integrity and the

Fig. 6. An overview of recommendations for integrating ChatGPT into an educational setting.

15
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

potential for biased or inappropriate content. Therefore, educators must the appropriate use of ChatGPT in various settings. By providing well-
adopt a proactive approach to ensure students know the potential risks reasoned explanations for each rule, teachers can foster a sense of
and ethical implications of using AI tools such as ChatGPT. This involves cooperation and understanding among students (McCabe, Trevino, &
discussing appropriate citation practices to avoid plagiarism, under­ Butterfield, 2001). For instance, a rule might state: “Students are
standing the potential ramifications of sharing personal or sensitive in­ encouraged to use ChatGPT as a supplementary resource to support their
formation with AI systems and informing users of their rights and learning and research activities, complementing their engagement with pri­
responsibilities regarding data ownership and usage. By doing so, mary sources, diverse perspectives and critical thinking skills, but not as a
learners are empowered to make informed decisions about utilizing and substitute for these essential components of learning. This approach ensures
protecting their information. that students do not excessively rely on AI and, in turn, get the space to
develop robust cognitive skills necessary for independent learning.”
5.2. Reimagining learning goals and objectives in the AI era By establishing well-defined parameters, educators can ensure that
students harness the potential of AI technologies while minimizing the
In the era of AI, educators must reconsider learning goals and ob­ risks of misuse or overreliance. Well-established psychological theories,
jectives, given the likelihood that students will use AI-assisted tools such such as the Theory of Planned Behaviour and the Self-Determination
as ChatGPT for learning. New objectives should focus on developing Theory, can support the establishment of these rules and guidelines.
students’ abilities to collaborate with AI systems, compare AI responses, The Theory of Planned Behaviour (TPB) posits that an individual’s
and think critically about the information provided. Activities encour­ intention to perform a specific behavior is determined by their attitude
aging students to debate with ChatGPT or analyze its in-depth responses towards the behavior, subjective norms and perceived behavioral con­
could be valuable. Additionally, educators must assess the relevance and trol (Ajzen, 1991). Educators can influence students’ attitudes and
meaningfulness of their courses and lessons in an AI-infused learning subjective norms regarding the responsible use of AI tools such as
environment. ChatGPT by defining clear rules and guidelines (especially considering
As students increasingly rely on AI tools for support, educators must the nascent stage of this technology). This approach can, in turn, shape
ensure that learning outcomes harmonize with this evolving landscape. students’ intentions to use technology responsibly and effectively.
Theoretical frameworks such as active learning (Prince, 2004), Furthermore, by explaining the rationale behind each rule, educators
constructivist learning theories (Scheer & Noweski, 2012), and Bloom’s can foster a sense of shared understanding and cooperation, strength­
Taxonomy (Forehand, 2010) offer valuable guidance for adapting ening the subjective norms that govern responsible AI tool usage.
learning goals and objectives to account for the presence of AI. Self-Determination Theory (SDT) is another psychological frame­
Bloom’s Taxonomy is a renowned learning framework for organizing work that emphasizes the importance of intrinsic motivation and au­
educational objectives into hierarchical levels of complexity and speci­ tonomy in human behavior (Tyack & Mekler, 2020). According to SDT,
ficity. The taxonomy comprises the following six levels: knowledge, individuals are likelier to engage in a behavior when they perceive it to
comprehension, application, analysis, synthesis and evaluation (Fore­ be autonomously chosen and intrinsically motivating. Establishing clear
hand, 2010). By aligning learning goals and objectives with Bloom’s rules and guidelines for AI tool usage and providing well-reasoned ex­
Taxonomy, educators can establish a structured approach to promoting planations can promote a sense of autonomy and intrinsic motivation
cognitive development and critical thinking skills in AI-enhanced among students. Students may be more inclined to follow the guidelines
learning. Integrating AI tools such as ChatGPT into the learning pro­ and adopt responsible behaviors once they understand the purpose and
cess can support objectives across the various levels of Bloom’s benefits of responsible AI usage.
Taxonomy: In addition to providing clear rules, maintaining open communica­
tion with students about the responsible use of AI tools is crucial. Reg­
⋅ Acquiring Factual Knowledge: ChatGPT can help students establish ular discussions on the potential benefits and pitfalls of using ChatGPT
foundational knowledge in various subjects. However, educators in learning can promote a deeper understanding of technology’s role in
must instruct students on effective research methods and encourage education. This dialogue can also facilitate the exchange of ideas,
using multiple sources for accurate understanding. feedback and experiences, enabling educators and students to collabo­
⋅ Understanding Complex Ideas: ChatGPT can provide explanations, ratively refine best practices for effectively integrating AI tools into the
examples and analogies to support the development of comprehen­ learning process.
sion skills. Herein, educators must assess and supplement AI- Moreover, educators should consider regularly reviewing and
generated explanations for a holistic understanding (without updating the rules and guidelines to safeguard their continued relevance
oversimplification). and effectiveness. As AI technologies evolve and new insights emerge
⋅ Applying Knowledge to Solve Problems: ChatGPT can assist students in from research and practice, adapting the guidelines to reflect these
problem-solving, brainstorming and generating ideas in diverse changes will be essential for maintaining an up-to-date and responsible
subjects, fostering enhanced problem-solving skills and concept approach to AI integration in education.
application.
⋅ Critically Analyzing AI-Generated Content: Educators must encourage 5.4. Embracing new evaluation approaches with AI integration
students to dissect AI-generated output, focusing on structure, as­
sumptions and biases, to develop critical thinking skills. Focusing on actual learning outcomes rather than solely on the ar­
⋅ Synthesizing Diverse Information: ChatGPT can help students combine tifacts produced by students is vital for outlining effective evaluation
information from different sources, promoting well-informed, approaches (Hadi Mogavi et al., 2022). In line with the extant literature,
balanced perspectives and a deeper understanding of knowledge our findings show that most educators concur that continuous assess­
interconnectedness. ment throughout the semester facilitates a more precise evaluation of
⋅ Evaluating AI-Generated Content: By engaging in debates with AI tools learning compared to a few high-stakes exams (Nistal, 2013). With the
or analysing AI-generated content in depth, students can develop advent of ChatGPT, educators are increasingly inclined to assess stu­
their evaluative reasoning skills, learning to assess the quality and dents primarily within classroom settings.
credibility of information and make informed judgments. Project-based assessments, which align with higher levels of thinking
in Bloom’s Taxonomy, are an efficient means of evaluating learning in
5.3. Developing and communicating rules for AI in education environments employing AI tools. These assessments emphasize the
application of knowledge and skills in real-world contexts, focusing on
Educators must develop and communicate specific rules outlining the development of higher-order cognitive processes such as analysis,

16
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

evaluation and creation. By devising and executing projects that demand contexts.
critical thinking, problem-solving and innovative thinking, educators
can motivate students to engage with AI tools responsibly and 6. Limitations
thoughtfully, fostering the growth of crucial twenty-first-century
competencies. Inherent to any research, our work is subject to certain limitations
Moreover, incorporating peer and self-assessment techniques can that must be acknowledged when evaluating its explanatory power and
enrich the evaluation process by promoting collaboration, reflection and generalizability. While our study provides valuable insights into the
metacognitive skills among students (Noonan & Duncan, 2005). These experiences of early ChatGPT adopters, it encounters typical challenges
methods empower students to assess their and their peers’ works, associated with qualitative research, such as unintentional researcher
providing opportunities for constructive feedback and fostering a deeper biases (also refer to Hadi Mogavi et al., 2022) and the complexities of
comprehension of the assessment criteria. Nonetheless, it is crucial to generalizing findings beyond the specific context of the study (see
acknowledge that most instructors deem focused guidance from a su­ Boustani, Tally, Kim, & Nippert-Eng, 2020; Noman, Das, & Patil, 2019;
pervisor indispensable in such activities. Such guidance enables students Hadi Mogavi et al., 2022 ).
to reflect more swiftly on their tasks, assignments and exams by A primary limitation concerns the narrow scope and scale of data
participating in peer and self-assessment activities and utilizing AI tools collection, which focuses on a particular subset of posts, time frames,
such as ChatGPT. Many educators argue that this approach results in and social media platforms. Consequently, our investigation may not
more meaningful and enduring learning outcomes. completely capture the myriad perspectives and experiences of ChatGPT
In addition to re-envisioning assessment and evaluation approaches, users, necessitating a prudent interpretation of our findings when
educators must address the ethical implications of AI integration within drawing broader conclusions about the larger population of early
learning environments. This entails confronting data privacy, algo­ adopters.
rithmic fairness and potential biases in AI-generated content. Integrating Another limitation to acknowledge is that our study does not purport
discussions of these ethical considerations into the curriculum and to encompass the perspectives of every demographic within the early
assessment practices enables educators to raise awareness and instill a ChatGPT adopters in education. Instead, it represents the opinions of a
sense of social responsibility among students when using AI tools for specific group of English-speaking users who have chosen to share their
assignments and exams. experiences on the social media platforms investigated in this research.
This nuance is vital, as not all users may be inclined to discuss their
5.5. Designing effective prompts and incorporating reflective practices in experiences and viewpoints online (see Huang, Jiang, & Wu, 2020; Hadi
AI-assisted education Mogavi et al., 2022).
To address some of these limitations and advance our understanding
The successful integration of AI tools such as ChatGPT in educational of user experiences and viewpoints, future research could employ a more
settings increasingly relies on crafting well-structured prompts and comprehensive methodological approach. By incorporating techniques
incorporating reflective practices. Thoughtful prompt design enables such as Delphi studies, interviews, and surveys, researchers can gather
educators to leverage ChatGPT’s potential to support instructional goals data from a larger and more diverse sample, thus providing a more
and enhance student learning. Furthermore, integrating reflective robust representation of the perspectives and experiences of early
practices fosters critical thinking about learning, using AI tools, and ChatGPT adopters in education.
developing essential skills and competencies.
Designing quality prompts begins with identifying and aligning clear 7. Future work
learning objectives with appropriate AI-supported activities. This pro­
cess entails evaluating ChatGPT’s strengths and limitations and As the nascent field of ChatGPT studies progresses, there is a wealth
considering students’ needs and contexts. Aligning prompts with of research opportunities that warrant further exploration to either
learning goals optimizes AI tools’ potential to support and enrich the corroborate or challenge the findings of our preliminary investigations.
learning experience. Based on the data collected and scrutinized in our research on the use of
Ensuring the accuracy, relevance and suitability of ChatGPT- ChatGPT in education on various social media platforms, we highlight
generated content requires testing prompts and verifying content. Edu­ several avenues for future scholarly inquiries.
cators must thoroughly review AI-generated content and cross-check it Regarding ChatGPT’s practical performance, it is also crucial to
with credible sources to maintain quality and reliability. Encouraging assess its effectiveness in relation to specific learning outcomes, such as
students to cite ChatGPT when using its content promotes accountability knowledge retention, skill development, and other pertinent objectives.
and ethical practices in AI tool usage. Controlled experiments can be employed to facilitate this evaluation and
Educators can facilitate reflection by prompting students to consider determine ChatGPT’s overall efficacy within diverse and specialized
the following questions: educational contexts. Concurrently, examining interpersonal dynamics
is essential to gauge the impact of ChatGPT on teacher-student and
⋅ How did ChatGPT contribute to my understanding of the topic? student-student relationships, interaction patterns, and communication
⋅ What strategies did I employ to verify the accuracy and reliability of modes. This research area should explore ChatGPT’s potential to pro­
the AI-generated content? mote peer learning and foster collaborative educational environments.
⋅ How can I use AI tools responsibly and ethically to support my A critical perspective entails examining the longitudinal effects of
learning? ChatGPT on learning outcomes and users’ perceptions. In-depth, longi­
⋅ What challenges did I encounter while using ChatGPT, and how can I tudinal investigations can provide a holistic understanding of ChatGPT’s
overcome them in the future? enduring influence on education. Concurrently, a comparative analysis
should be undertaken to evaluate ChatGPT against other AI chatbots and
Lastly, educators should present ChatGPT-generated content engag­ traditional pedagogical methods. This comparative framework will
ingly, incorporating sound pedagogy and study approaches. Techniques enable a comprehensive examination of the unique advantages and
such as challenging questions, problem-solving, providing explanations challenges associated with ChatGPT and alternative educational
and storytelling foster active engagement, critical thinking and knowl­ approaches.
edge retention. For instance, educators can use ChatGPT-generated The ethical and responsible integration of ChatGPT in educational
content to create case studies, simulations or role-playing activities, settings necessitates an examination of policy development and teacher
allowing students to apply their knowledge and skills in real-world training. Policymakers, including academic institutions, must devise and

17
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

implement regulations to ensure the judicious use of ChatGPT. Ahn, M. J., & Chen, Y.-C. (2020). Artificial intelligence in government: Potentials,
challenges, and the future. In The 21st annual international conference on digital
Furthermore, it is crucial to evaluate strategies for equipping educators
government research, dg.o ’20 (pp. 243–252). New York, NY, USA: Association for
with the necessary skills and knowledge to seamlessly incorporate Computing Machinery. https://doi.org/10.1145/3396956.3398260, 10.1145/
ChatGPT into their teaching practices. 3396956.3398260.
Addressing the varied needs of learners, research should investigate Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human
Decision Processes, 50(2), 179–211.
ChatGPT’s ability to cater to culturally and linguistically diverse stu­ Aktan, M. E., Turhan, Z., & Dolu, İ. (2022). Attitudes and perspectives towards the
dents, as well as its application in special education. This line of inquiry preferences for artificial intelligence in psychotherapy. Computers in Human Behavior,
should evaluate potential synergies between ChatGPT and other assis­ 133, Article 107273. https://doi.org/10.1016/j.chb.2022.107273, 10.1016/j.
chb.2022.107273.
tive technologies to support students with special educational needs, Ao, S., Junchi, W., Ke, L., Yuxin, S., Bingxiang, Y., & Zhisheng, H. (2021). Analysis of
learning disabilities, or other distinct learning challenges. needs and availability of information on depression based on the zhihu q&a
Finally, it is essential to explore innovative approaches to enrich the community. In Proceedings of the 2nd international symposium on artificial intelligence
for medicine sciences (pp. 551–556). New York, NY, USA: ISAIMS ’21, Association for
learning experience with ChatGPT. One such approach involves utilizing Computing Machinery. https://doi.org/10.1145/3500931.3501025, 10.1145/
affective computing to assess ChatGPT’s ability to identify and respond 3500931.3501025.
to the emotional and social dimensions of learning, thereby catering to Baktha, K., & Tripathy, B. K. (2017). Investigation of recurrent neural networks in the
field of sentiment analysis. In 2017 international conference on communication and
learners’ affective needs. Additionally, integrating ChatGPT with signal processing (ICCSP). IEEE. https://doi.org/10.1109/iccsp.2017.8286763,
various instructional strategies, such as game-based learning, gamifi­ 10.1109/iccsp.2017.8286763.
cation techniques, blended learning, and multimodal content delivery, Beltagy, I., Cohan, A., Logan IV, R., Min, S., & Singh, S. (2022). Zero-and few-shot nlp
with pretrained language models. In Proceedings of the 60th annual meeting of the
can serve to enhance student motivation, engagement, and learning
association for computational linguistics: Tutorial abstracts (pp. 32–37).
outcomes. Bianchi, T. (2023). Global weekly interest in chatgpt on google search 2022-2023. https
Moreover, the potential of combining ChatGPT with metaverse- ://www.statista.com/statistics/1366930/chatgpt-google-search-weekly-worldwide
based classrooms should be examined. Metaverse-based classrooms /?locale=en.
Blandford, A., Furniss, D., & Makri, S. (2016). Qualitative HCI research. Springer
(see Zhou, 2023), utilizing immersive virtual environments, offer op­ International Publishing. https://doi.org/10.1007/978-3-031-02217-3, 10.1007/
portunities for transformative pedagogical experiences that foster 978-3-031-02217-3.
deeper learning and collaboration. Investigating the synergy between Boustani, K., Tally, A. C., Kim, Y. R., & Nippert-Eng, C. (2020). Gaming the name: Player
strategies for adapting to name constraints in online videogames. In Proceedings of
ChatGPT and metaverse-enhanced learning environments can provide the annual symposium on computer-human interaction in play, CHI PLAY ’20 (pp.
valuable insights into the effectiveness of these innovative approaches in 120–131). New York, NY, USA: Association for Computing Machinery. https://doi.
promoting a more engaging and enriching educational experience for org/10.1145/3410404.3414259, 10.1145/3410404.3414259.
Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative
learners. Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/
1478088706qp063oa, 10.1191/1478088706qp063oa.
Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., et al. (2020).
8. Conclusion
Language models are few-shot learners. In H. Larochelle, M. Ranzato, R. Hadsell,
M. Balcan, & H. Lin (Eds.), Advances in neural information processing systems (Vol. 33,
In this study, we investigated the adoption and perception of pp. 1877–1901). Curran Associates, Inc.. URL https://proceedings.neurips.cc/pape
ChatGPT in education by analyzing qualitative data collected from r_files/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf.
Calisto, F. M., Fernandes, J., Morais, M., Santiago, C., Abrantes, J. M., Nunes, N., et al.
various social media platforms. Our analysis revealed that ChatGPT is (2023). Assertiveness-based agent communication for a personalized medicine on
utilized in diverse settings for multiple purposes, with the most wide­ medical imaging diagnosis. In Proceedings of the 2023 CHI conference on human factors
spread uses observed in higher education, K-12 education, and practical in computing systems. ACM. https://doi.org/10.1145/3544548.3580682, 10.1145/
3544548.3580682.
skills development. Calisto, F. M., Nunes, N., & Nascimento, J. C. (2022). Modeling adoption of intelligent
Among the numerous applications of ChatGPT in higher education, agents in medical imaging. International Journal of Human-Computer Studies, 168,
content creation and editing emerged as the most prevalent, accounting Article 102922. https://doi.org/10.1016/j.ijhcs.2022.102922, 10.1016/j.
ijhcs.2022.102922.
for 78.11% of reported uses. Our findings unveiled a range of promising Calisto, F. M., Santiago, C., Nunes, N., & Nascimento, J. C. (2021). Introduction of
opportunities for ChatGPT to support learning, while simultaneously human-centric AI assistant to aid radiologists for multimodal breast image
uncovering critical risks associated with overreliance on the tool. These classification. International Journal of Human-Computer Studies, 150, Article 102607.
https://doi.org/10.1016/j.ijhcs.2021.102607, 10.1016/j.ijhcs.2021.102607.
risks encompass the potential to limit critical thinking and creativity,
Calisto, F. M., Santiago, C., Nunes, N., & Nascimento, J. C. (2022). BreastScreening-AI:
impede a deep understanding of subject matter, and foster laziness and Evaluating medical intelligent agents for human-AI interactions. Artificial Intelligence
passivity. in Medicine, 127, Article 102285. https://doi.org/10.1016/j.artmed.2022.102285,
10.1016/j.artmed.2022.102285.
Drawing upon insights gleaned from the abundance of crowdsourced
Cheng, K.-H., & Tsai, C.-C. (2014). The interaction of child-parent shared reading with an
knowledge on social media platforms, our study offered recommenda­ augmented reality (AR) picture book and parents’ conceptions of AR learning. British
tions for a more responsible and effective integration of ChatGPT in Journal of Educational Technology, 47(1), 203–222. https://doi.org/10.1111/
educational contexts. Moving forward, it is clear that ongoing research is bjet.12228, 10.1111/bjet.12228.
Chen, Y., Wang, K.-R., Xu, W., & Huang, Y. (2021). Exploring commenting behavior in
crucial for tracking the progress of language models and AI tools like the covid-19 super-topic on weibo. In Extended abstracts of the 2021 CHI conference on
ChatGPT, as well as assessing their influence on learning outcomes. In human factors in computing systems, CHI EA ’21. New York, NY, USA: Association for
accordance with this perspective, we concluded our paper by outlining Computing Machinery. https://doi.org/10.1145/3411763.3451717, 10.1145/
3411763.3451717.
an extensive research agenda for future exploration. Chetlen, A., Artrip, R., Drury, B., Arbaiza, A., & Moore, M. (2019). Novel use of chatbot
technology to educate patients before breast biopsy. Journal of the American College
of Radiology, 16(9), 1305–1308. https://doi.org/10.1016/j.jacr.2019.05.050,
Declaration of competing interest 10.1016/j.jacr.2019.05.050.
Clarizia, F., Colace, F., Lombardi, M., Pascale, F., & Santaniello, D. (2018). Chatbot: An
education support system for student. In Cyberspace safety and security (pp.
The authors declare that they have no known competing financial 291–302). Springer International Publishing. https://doi.org/10.1007/978-3-030-
interests or personal relationships that could have appeared to influence 01689-0_23, 10.1007/978-3-030-01689-0_23.
the work reported in this paper. Conroy, S.. What countries is chat gpt available & not available in? (April 2023). https://
www.wepc.com/tips/what-countries-is-chat-gpt-unavailable/.
Dale, R. (2020). GPT-3: What’s it good for? Natural Language Engineering, 27(1),
References 113–118. https://doi.org/10.1017/s1351324920000601, 10.1017/
s1351324920000601.
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). Bert: Pre-training of deep
Abdullah, M., Madain, A., & Jararweh, Y. (2022). ChatGPT: Fundamentals, applications
bidirectional transformers for language understanding. arXiv preprint arXiv:
and social impacts. In 2022 ninth international conference on social networks analysis,
1810.04805.
management and security (SNAMS). IEEE. https://doi.org/10.1109/
snams58071.2022.10062688, 10.1109/snams58071.2022.10062688.

18
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

Dong, L., Yang, N., Wang, W., Wei, F., Liu, X., Wang, Y., et al. (2019). Unified language Khosravi, H., Shum, S. B., Chen, G., Conati, C., Tsai, Y.-S., Kay, J., et al. (2022).
model pre-training for natural language understanding and generation. In Explainable artificial intelligence in education. Computers & Education: Artificial
H. Wallach, H. Larochelle, A. Beygelzimer, F. d’Alché-Buc, E. Fox, & R. Garnett Intelligence, 3, Article 100074. https://doi.org/10.1016/j.caeai.2022.100074,
(Eds.), Advances in neural information processing systems (Vol. 32). Curran Associates, 10.1016/j.caeai.2022.100074.
Inc.. https://proceedings.neurips.cc/paper_files/paper/2019/file/c20bb2d9a50d5 Kikerpill, K., & Siibak, A. (2023). App-hazard disruption: An empirical investigation of
ac1f713f8b34d9aac5a-Paper.pdf media discourses on ChatGPT in educational contexts. Computers in the Schools, 1–22.
Elsaid, A., Mohammed, A., Ibrahim, L. F., & Sakre, M. M. (2022). A comprehensive https://doi.org/10.1080/07380569.2023.2244941, 10.1080/
review of Arabic text summarization. IEEE Access, 10, 38012–38030. https://doi. 07380569.2023.2244941.
org/10.1109/access.2022.3163292, 10.1109/access.2022.3163292. Kulkarni, A., Shivananda, A., & Kulkarni, A. (2021). Building a chatbot using transfer
Esmaeilzadeh, A., & Taghva, K. (2021). Text classification using neural network language learning. In Natural Language processing projects (pp. 239–255). Apress. https://doi.
model (NNLM) and BERT: An empirical comparison. In Lecture notes in networks and org/10.1007/978-1-4842-7386-9_9, 10.1007/978-1-4842-7386-9_9.
systems (pp. 175–189). Springer International Publishing. https://doi.org/10.1007/ Lauriola, I., Lavelli, A., & Aiolli, F. (2022). An introduction to deep learning in natural
978-3-030-82199-9_12, 10.1007/978-3-030-82199-9_12. language processing: Models, techniques, and tools. Neurocomputing, 470, 443–456.
AoIR. (2019). Association of internet researchers (AoIR), internet research: Ethical guidelines https://doi.org/10.1016/j.neucom.2021.05.103, 10.1016/j.neucom.2021.05.103.
3.0. https://aoir.org/ethics/. Liang, Q., de la Torre, J., & Law, N. (2021). Do background characteristics matter in
2023 P, Zhou. (2023). Unleasing chatgpt on the metaverse: Savior or destroyer? arXiv children’s mastery of digital literacy? A cognitive diagnosis model analysis.
preprint arXiv:2303.13856. Computers in Human Behavior, 122, Article 106850. https://doi.org/10.1016/j.
Bang, Y., Cahyawijaya, S., Lee, N., Dai, W., Su, D., Wilie, B., et al. (2023). A multitask, chb.2021.106850, 10.1016/j.chb.2021.106850.
multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and J. Jeon, S. Lee. (2023). Large language models in education: A focus on the
interactivity. arXiv preprint arXiv:2302.04023. complementary relationship between human teachers and ChatGPT, Education and
Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., et al. Information Technologiesdoi:10.1007/s10639-023-11834-1 URL https://doi.
(2023). Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv org/10.1007/s10639-023-11834-1.
preprint arXiv:2303.12712. M. Lewis, Y. Liu, N. Goyal, M. Ghazvininejad, A. Mohamed, O. Levy, V. Stoyanov, L.
Forehand, M. (2010). Bloom’s taxonomy, Emerging perspectives on learning, teaching, and Zettlemoyer. (2019). Bart: Denoising sequence-to-sequence pre-training for natural
technology, 41(4), 47–56. language generation, translation, and comprehension, arXiv preprint arXiv:
Foroughi, B., Senali, M. G., Iranmanesh, M., Khanfar, A., Ghobakhloo, M., Annamalai, N., 1910.13461.
et al. (2023). Determinants of intention to use ChatGPT for educational purposes: Lin, B., & Huang, E. M. (2010). Reuse: Promoting repurposing through an online diy
Findings from PLS-SEM and fsQCA. International Journal of Human-Computer community. In CHI ’10 extended abstracts on human factors in computing systems, CHI
Interaction, 1–20. https://doi.org/10.1080/10447318.2023.2226495, 10.1080/ EA ’10 (pp. 4177–4182). New York, NY, USA: Association for Computing Machinery.
10447318.2023.2226495. https://doi.org/10.1145/1753846.1754122, 10.1145/1753846.1754122.
GDPR. (2018). The European Union’s general data protection regulation. https://gdpr-info. LinkedIn, I.. Linkedin api products (January 2023). https://developer.linkedin.com/pr
eu/. oduct-catalog.
Atlas. (2023). A computer-assisted qualitative data analysis software. https://atlasti. Li, J., Shang, S., & Shao, L. (2020). Metaner: Named entity recognition with meta-
com. learning. In Proceedings of the web conference 2020 (pp. 429–440). New York, NY,
Gong, J., Yao, Z., Lu, Z., Ding, Q., zhang, y., Zhang, L., et al. (2021). All in one group: USA: WWW ’20, Association for Computing Machinery. https://doi.org/10.1145/
Current practices, lessons and challenges of Chinese home-school communication in 3366423.3380127, 10.1145/3366423.3380127.
im group chat. In Proceedings of the 2021 CHI conference on human factors in computing Liu, J., Loh, L., Ng, E., Chen, Y., Wood, K. L., & Lim, K. H. (2020). Self-evolving adaptive
systems, CHI ’21. New York, NY, USA: Association for Computing Machinery. https:// learning for personalized education. In Conference companion publication of the 2020
doi.org/10.1145/3411764.3445436, 10.1145/3411764.3445436. on computer supported cooperative work and social computing, CSCW ’20 companion
Google, I.. Youtube data api (March 2023). https://developers.google.com/youtube/v3. (pp. 317–321). New York, NY, USA: Association for Computing Machinery. https://
Hadi Mogavi, R., Guo, B., Zhang, Y., Haq, E.-U., Hui, P., & Ma, X. (2022). When doi.org/10.1145/3406865.3418326, 10.1145/3406865.3418326.
gamification spoils your learning: A qualitative case study of gamification misuse in Luo, W., He, H., Liu, J., Berson, I. R., Berson, M. J., Zhou, Y., et al. (2023). Aladdin’s genie
a language-learning app. In Proceedings of the ninth ACM conference on learning @ or pandora’s box for early childhood education? Experts chat on the roles, challenges, and
scale (pp. 175–188). New York, NY, USA: L@S ’22, Association for Computing developments of ChatGPT. Early Education and Development. https://doi.org/
Machinery. https://doi.org/10.1145/3491140.3528274, 10.1145/ 10.1080/10409289.2023.2214181, 10.1080/10409289.2023.2214181.
3491140.3528274. McCabe, D. L., Trevino, L. K., & Butterfield, K. D. (2001). Cheating in academic
R. Fogliato, A. Chouldechova, Z. Lipton. (2021). The impact of algorithmic risk institutions: A decade of research. Ethics & Behavior, 11(3), 219–232. https://doi.
assessments on human predictions and its analysis via crowdsourcing studies, Proc. org/10.1207/s15327019eb1103_2, 10.1207/s15327019eb1103_2.
ACM Hum.-Comput. Interact. 5 (CSCW2). doi:10.1145/3479572 URL https://doi. Mogavi, R. H., Ma, X., & Hui, P. (2021). Characterizing student engagement moods for
org/10.1145/3479572. dropout prediction in question pool websites. Proceedings of the ACM on Human-
Hadi Mogavi, R., Zhao, Y., Ul Haq, E., Hui, P., & Ma, X. (2021). Student barriers to active Computer Interaction, 5(CSCW1), 1–22. https://doi.org/10.1145/3449086, 10.1145/
learning in synchronous online classes: Characterization, reflections, and 3449086.
suggestions. In Proceedings of the eighth ACM conference on learning @ scale, L@S ’21, Muller, M., Chilton, L. B., Kantosalo, A., Martin, C. P., & Walsh, G. (2022). Genaichi:
(pp. 101–115). New York, NY, USA: Association for Computing Machinery. https:// Generative ai and hci. In Extended abstracts of the 2022 CHI conference on human
doi.org/10.1145/3430895.3460126, 10.1145/3430895.3460126. factors in computing systems, CHI EA ’22. New York, NY, USA: Association for
Hao, K. (2019). Openai has released the largest version yet of its fake-news-spewing ai. Computing Machinery. https://doi.org/10.1145/3491101.3503719, 10.1145/
https://www.technologyreview.com/2019/08/29/133218/openai-released-its- 3491101.3503719.
fake-news-ai-gpt-2/. Murgia, E., Pera, M. S., Landoni, M., & Huibers, T. (2023). Children on chatgpt
Haq, E.-U., Braud, T., Lee, L.-H., Vallapuram, A. K., Yu, Y., Tyson, G., et al. (2022). Short, readability in an educational context: Myth or opportunity?. In Adjunct proceedings of
colorful, and irreverent! a comparative analysis of new users on wallstreetbets the 31st ACM conference on user modeling, adaptation and personalization, UMAP ’23
during the gamestop short-squeeze. In Companion proceedings of the web conference adjunct (pp. 311–316). New York, NY, USA: Association for Computing Machinery.
2022 (pp. 52–61). New York, NY, USA: WWW ’22, Association for Computing https://doi.org/10.1145/3563359.3596996, 10.1145/3563359.3596996.
Machinery. https://doi.org/10.1145/3487553.3524202, 10.1145/ Nistal, M. L. (2013). An experience of continuous assessment in telecommunication
3487553.3524202. technologies engineering: New costs for the teacher. IEEE Revista Iberoamericana de
Hoi, S. C. (2023). Responsible ai for trusted ai-powered enterprise platforms. In Tecnologias del Aprendizaje, 8(2), 90–95. https://doi.org/10.1109/
Proceedings of the sixteenth ACM international conference on web search and data rita.2013.2258225, 10.1109/rita.2013.2258225.
mining, WSDM ’23 (pp. 1277–1278). New York, NY, USA: Association for Computing Noman, A. S. M., Das, S., & Patil, S. (2019). Techies against facebook: Understanding
Machinery. https://doi.org/10.1145/3539597.3575784, 10.1145/ negative sentiment toward facebook via user generated content. In Proceedings of the
3539597.3575784. 2019 CHI conference on human factors in computing systems, CHI ’19 (pp. 1–15). New
Holmes, W., & Tuomi, I. (2022). State of the art and practice in ai in education. European York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/
Journal of Education, 57(4), 542–570. https://doi.org/10.1111/ejed.12533, 3290605.3300698, 10.1145/3290605.3300698.
10.1111/ejed.12533. Noonan, B., & Duncan, C. R. (2005). Peer and self-assessment in high schools. Practical
R. Hadi Mogavi, Y. Zhang, E.-U. Haq, Y. Wu, P. Hui, X. Ma. (2022). What do users think Assessment, Research and Evaluation, 10(1), 17.
of promotional gamification schemes? a qualitative case study in a question Omran, T. M., Sharef, B. T., Grosan, C., & Li, Y. (2023). Transfer learning and sentiment
answering website, Proc. ACM Hum.-Comput. Interact. 6 (CSCW2).doi:10.1145/ analysis of bahraini dialects sequential text data using multilingual deep learning
3555124. URL https://doi.org/10.1145/3555124. approach. Data & Knowledge Engineering, 143, Article 102106. https://doi.org/
Hwang, G.-J., & Chang, C.-Y. (2021). A review of opportunities and challenges of 10.1016/j.datak.2022.102106, 10.1016/j.datak.2022.102106.
chatbots in education. Interactive Learning Environments, 1–14. https://doi.org/ OpenAI. (2023). Gpt-4 technical report. arXiv:2303.08774.
10.1080/10494820.2021.1952615, 10.1080/10494820.2021.1952615. X. Lin, (2023). Exploring the role of ChatGPT as a facilitator for motivating self-directed
C. Huang, W. Jiang, J. Wu, G. Wang. (2020). Personalized review recommendation based learning among adult learners, Adult Learningdoi:10.1177/10451595231184928
on users’ aspect sentiment, ACM Transactions on Internet Technology 20 (4). doi: URL https://doi.org/10.1177/10451595231184928.
10.1145/3414841 URL https://doi.org/10.1145/3414841. P. Liu, W. Yuan, J. Fu, Z. Jiang, H. Hayashi, G. Neubig, (2023) Pre-train, prompt, and
Kasthuri, E., & Balaji, S. (2021). A chatbot for changing lifestyle in education. In 2021 predict: A systematic survey of prompting methods in natural language processing,
third international conference on intelligent communication technologies and virtual ACM Computing Surveys 55 (9). doi:10.1145/3560815. URLhttps://doi.org/
mobile networks (ICICV). IEEE. https://doi.org/10.1109/icicv50876.2021.9388633, 10.1145/3560815.
10.1109/icicv50876.2021.9388633.

19
R. Hadi Mogavi et al. Computers in Human Behavior: Artificial Humans 2 (2024) 100027

Dwivedi, YK, Kshetri, N, Hughes, L, Slade, EL, Jeyaraj, A, Kar, AK, et al. (2023). Opinion Tyack, A., & Mekler, E. D. (2020). Self-determination theory in hci games research:
paper: “so what if ChatGPT wrote it?” multidisciplinary perspectives on Current uses and open questions. In Proceedings of the 2020 CHI conference on human
opportunities, challenges and implications of generative conversational AI for factors in computing systems, CHI ’20 (pp. 1–22). New York, NY, USA: Association for
research, practice and policy. International Journal of Information Management, 71, Computing Machinery. https://doi.org/10.1145/3313831.3376723, 10.1145/
Article 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642, 10.1016/j. 3313831.3376723.
ijinfomgt.2023.102642. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., et al. (2017).
Pater, J. A., Reining, L. E., Miller, A. D., Toscos, T., Mynatt, E. D., & notjustgirls”. (2019). Attention is all you need. arXiv:1706.03762 [cs], 10.48550/arXiv.1706.03762 htt
Exploring male-related eating disordered content across social media platforms. In p://arxiv.org/abs/1706.03762.
Proceedings of the 2019 CHI conference on human factors in computing systems, CHI ’19 Wang, C.-Y., & Lin, J. J. (2023). Utilizing artificial intelligence to support analyzing self-
(pp. 1–13). New York, NY, USA: Association for Computing Machinery. https://doi. regulated learning: A preliminary mixed-methods evaluation from a human-centered
org/10.1145/3290605.3300881, 10.1145/3290605.3300881. perspective. Computers in Human Behavior, 144, Article 107721. https://doi.org/
Prince, M. (2004). Does active learning work? A review of the research. Journal of 10.1016/j.chb.2023.107721, 10.1016/j.chb.2023.107721.
Engineering Education, 93(3), 223–231. H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, et al., (2023).
K. Papadamou, S. Zannettou, J. Blackburn, E. De Cristofaro, G. Stringhini, M. Sirivianos. Llama: Open and efficient foundation language models, arXiv preprint arXiv:
(2021). ”how over is it?” understanding the incel community on youtube, Proc. ACM 2302.13971.
Hum.-Comput. Interact. 5 (CSCW2). doi:10.1145/3479556. URLhttps://doi.org/ Wolf, T., Debut, L., Sanh, V., Chaumond, J., Delangue, C., Moi, A., et al. (2020).
10.1145/3479556. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020
A. Radford, K. Narasimhan, T. Salimans, I. Sutskever, et al. (2018). Improving language conference on empirical methods in Natural Language Processing: System demonstrations.
understanding by generative pre-training. Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.
A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever, et al. (2019). Language emnlp-demos.6, 10.18653/v1/2020.emnlp-demos.6.
models are unsupervised multitask learners. Xia, H., Ng, H. X., Chen, Z., & Hollan, J. (2022). Millions and billions of views:
Ray, P. P. (2023). ChatGPT: A comprehensive review on background, applications, key Understanding popular science and knowledge communication on video-sharing
challenges, bias, ethics, limitations and future scope. Internet of Things and Cyber- platforms. In Proceedings of the ninth ACM conference on learning @ scale (pp.
Physical Systems, 3, 121–154. https://doi.org/10.1016/j.iotcps.2023.04.003, 163–174). New York, NY, USA: L@S ’22, Association for Computing Machinery.
10.1016/j.iotcps.2023.04.003. https://doi.org/10.1145/3491140.3528279, 10.1145/3491140.3528279.
Reddit, I.. Reddit api (january 2023). https://www.reddit.com/dev/api/. Xia, M., Sun, M., Wei, H., Chen, Q., Wang, Y., Shi, L., et al. (2019). Peerlens: Peer-
Ruby, D.. Chatgpt statistics for 2023 (new data + gpt-4 facts) (March 2023). URL htt inspired interactive learning path planning in online question pool. In Proceedings of
ps://www.demandsage.com/chatgpt-statistics/. the 2019 CHI conference on human factors in computing systems, CHI ’19 (pp. 1–12).
Ruby, D.. Reddit statistics for 2023 (users & traffic data) (March 2023). https://www. New York, NY, USA: Association for Computing Machinery. https://doi.org/
demandsage.com/. 10.1145/3290605.3300864, 10.1145/3290605.3300864.
Ruby, D.. Youtube statistics (2023) — trending facts & figures shared! (January 2023). Yang, W. (2022). Artificial intelligence education for young children: Why, what, and
https://www.demandsage.com/. how in curriculum design and implementation. Computers & Education: Artificial
Ruby, D. (2023a). Twitter statistics for marketers in 2023 (users & trends). https://www. Intelligence, 3, Article 100061. https://doi.org/10.1016/j.caeai.2022.100061,
demandsage.com/. 10.1016/j.caeai.2022.100061.
Ruby, D. (2023b). Important linkedin statistics for 2023 (data & trends). https://www. Yang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R. R., & Le, Q. V. (2019). Xlnet:
demandsage.com/. Generalized autoregressive pretraining for language understanding. In H. Wallach,
J. W. Rae, S. Borgeaud, T. Cai, K. Millican, J. Hoffmann, F. Song, J. Aslanides, S. H. Larochelle, A. Beygelzimer, F. d’Alché-Buc, E. Fox, & R. Garnett (Eds.), Advances
Henderson, R. Ring, S. Young, et al., (2021). Scaling language models: Methods, in neural information processing systems (Vol. 32). Curran Associates, Inc.. https
analysis & insights from training gopher, arXiv preprint arXiv:2112.11446. ://proceedings.neurips.cc/paper_files/paper/2019/file/dc6a7e655d7e5840e66733e
Seldon, A., & Abidoye, O. (2018). The fourth education revolution. Legend Press Ltd. 9ee67cc69-Paper.pdf
Shoufan, A. (2023). Exploring students’ perceptions of ChatGPT: Thematic analysis and Yang, S., Wang, Y., & Chu, X. (2020). A survey of deep learning techniques for neural
follow-up survey. IEEE Access, 11, 38805–38818. https://doi.org/10.1109/ machine translation, 10.48550/ARXIV.2002.07526 https://arxiv.org/abs/2002.0752
access.2023.3268224, 10.1109/access.2023.3268224. 6.
Sinha, R., Shibata, R., Patel, A., & Sternchos, J. A. (2021). Social media in minimally Yao, L., & Guan, Y. (2018). An improved LSTM structure for natural language processing.
invasive gynecologic surgery: What is #trending on instagram? Journal of Minimally In 2018 IEEE international conference of safety produce informatization (IICSPI). IEEE.
Invasive Gynecology, 28(10), 1730–1734. https://doi.org/10.1016/j. https://doi.org/10.1109/iicspi.2018.8690387, 10.1109/iicspi.2018.8690387.
jmig.2021.02.011, 10.1016/j.jmig.2021.02.011. Yi-No Kang, E., Chen, D.-R., & Chen, Y.-Y. (2023). Associations between literacy and
Smith, C. P., Atkinson, J. W., McClelland, D. C., Veroff, J., et al. (1992). Motivation and attitudes toward artificial intelligence–assisted medical consultations: The mediating
personality: Handbook of thematic content analysis. Cambridge University Press. role of perceived distrust and efficiency of artificial intelligence. Computers in Human
A. Scheer, C. Noweski, C. Meinel. (2012). Transforming constructivist learning into Behavior, 139, Article 107529. https://doi.org/10.1016/j.chb.2022.107529. https://
action: Design thinking in education, Design and Technology Education: www.sciencedirect.com/science/article/pii/S0747563222003491
International Journal 17 (3). Zhai, X., Chu, X., Chai, C. S., Jong, M. S. Y., Istenic, A., Spector, M., et al. (2021).
Tahaei, M., Vaniea, K., & Saphra, N. (2020). Understanding privacy-related questions on A review of artificial intelligence (AI) in education from 2010 to 2020. Complexity,
stack overflow. In Proceedings of the 2020 CHI conference on human factors in 2021, 1–18. https://doi.org/10.1155/2021/8812542, 10.1155/2021/8812542.
computing systems, CHI ’20 (pp. 1–14). New York, NY, USA: Association for Zhang, A. (2022). Human computer interaction system for teacher-student interaction
Computing Machinery. https://doi.org/10.1145/3313831.3376768, 10.1145/ model using machine learning. International Journal of Human-Computer Interaction,
3313831.3376768. 1–12. https://doi.org/10.1080/10447318.2022.2115645, 10.1080/
S. Sok, K. Heng. (2023). Chatgpt for education and research: A review of benefits and 10447318.2022.2115645.
risks, Available at SSRN 4378735. Zhang, L., Wang, S., & Liu, B. (2018). Deep learning for sentiment analysis: A survey,
H. Tinmaz, Y.-T. Lee, M. Fanea-Ivanovici, H. Baber. (2022). A systematic review on WIREs Data Mining and Knowledge. Discovery, 8((4). doi:10.1002/widm.1253URL).
digital literacy, Smart Learning Environments 9 (1). doi:10.1186/s40561-022- https://doi.org/10.1002/widm.1253.
00204-y URL https://doi.org/10.1186/s40561-022-00204-y. Zhong, N., Qian, Z., & Zhang, X. (2021). Deep neural network retrieval. In Proceedings of
A. Tlili, R. Huang, B. Shehata, D. Liu, J. Zhao, A. H. S. Metwally, et al. (2022) Is the 29th ACM international conference on multimedia. ACM. https://doi.org/10.1145/
metaverse in education a blessing or a curse: a combined content and bibliometric 3474085.3475505, 10.1145/3474085.3475505.
analysis, Smart Learning Environments 9 (1). doi:10.1186/s40561-022-00205-x URL D. Wang, J. D. Weisz, M. Muller, P. Ram, W. Geyer, C. Dugan, et al. (2019). Human-ai
https://doi.org/10.1186/s40561-022-00205-x. collaboration in data science: Exploring data scientists’ perceptions of automated ai,
A. Tlili, B. Shehata, M. A. Adarkwah, A. Bozkurt, D. T. Hickey, R. Huang, B. Agyemang, Proc. ACM Hum.-Comput. Interact. 3 (CSCW). doi:10.1145/3359313 URL htt
(2023) What if the devil is my guardian angel: ChatGPT as a case study of using ps://doi.org/10.1145/3359313.
chatbots in education, Smart Learning Environments 10 (1). doi:10.1186/s40561- C. Zhang, C. Zhang, S. Zheng, Y. Qiao, C. Li, M. Zhang, et al., (2023). A complete survey
023-00237-x.URL https://doi.org/10.1186/s40561-023-00237-x. on generative ai (aigc): Is chatgpt from gpt-4 to gpt-5 all you need?, arXiv preprint
Twitter, I.. Twitter api (january 2023). https://developer.twitter.com/en/docs/twitte arXiv:2303.11717.
r-api.

20