Computers and Education: Artificial Intelligence 5 (2023) 100184

Contents lists available at ScienceDirect

Computers and Education: Artificial Intelligence

journal homepage: www.sciencedirect.com/journal/computers-and-education-artificial-intelligence

Integrating generative AI in knowledge building

Bodong Chen a, *, Xinran Zhu a, Fernando Díaz del Castillo H. b
a
University of Pennsylvania, Philadelphia, USA
b
Mentu and Gimnasio La Montaña, Bogotá, Colombia

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

Keywords: Generative artificial intelligence (GenAI) is penetrating in various social sectors, motivating a strong need for
Generative AI teaching AI literacy in younger generations. While substantial efforts have been made to teach AI literacy and to
ChatGPT use AI to facilitate learning, few studies have provided empirical accounts of students’ nuanced processes of
Knowledge building
using GenAI for learning. In this study, we engaged a group of high school students in leveraging ChatGPT to
Human-AI partnership
support their knowledge building efforts. Following the teacher’s pedagogical design, students used ChatGPT for
a range of distinct purposes. Student interviews showed detailed processes of using ChatGPT for knowledge
building and students’ emerging AI literacy in multiple dimensions. This study offers practical implications for
the integration of GenAI in K-12 education and urges educators to create spaces and scaffolds for students to
mindfully engage with GenAI in the classroom.

1. Introduction generate news content, leading to increased efficiency and productivity,

but also concerns about the quality of news and misinformation (Pavlik,
Artificial Intelligence (AI) is having a significant impact on industry, 2023). In education, while some educators have welcomed ChatGPT’s
education, entertainment, and virtually all sectors of society (Leander & potential for improving the learning experience for students, others have
Burriss, 2020). Broadly speaking, AI can be defined as the ability of raised concerns about its impact on academic integrity and plagiarism
machines to perform tasks that typically require human intelligence, (Kasneci et al., 2023; UNESCO, 2023). Broad attention attracted by
such as visual perception, speech recognition, decision-making, and ChatGPT has accelerated public discourse of the impact of GenAI, as well
language understanding (Russell & Norvig, 1995). AI tools often rely on as other types of AI, on society, spurring the development of safe­
statistical methods, machine learning models, and computer algorithms, guarding mechanisms necessary for guiding AI advancements towards
and are increasingly deployed in socio-technical systems such as content greater good. For educators, it becomes pivotal to create opportunities
filters on social media, autonomous vehicles, mortgage approval sys­ for learners to examine what AI is, how it works, and how it can be
tems, and chatbots—just to name a few. Given the increasingly broader meaningfully deployed to support human flourishing in this new era.
reach of AI in society, concerns have been raised in its roles in political Indeed, we have witnessed a sharply growing interest in introducing
polarization, algorithmic bias, proliferation of surveillance, etc., K-12 students to AI in recent years. Because AI is poised to transform
attracting critical engagement by scholars from a wide range of disci­ many industries, by learning about AI early on, students can develop the
plines including political science, sociology, media studies, environ­ knowledge and skills necessary to thrive in this changing job market.
mental science, geography, medicine, and education. Even more importantly, since AI is becoming increasingly prevalent in
Recent advances in generative artificial intelligence (GenAI) and everyday life, understanding how AI works and its potential applications
large language models (LLMs) have attracted even broader societal can help students become “AI-literate” citizens who are better equipped
attention (Leander & Burriss, 2020). ChatGPT in particular has caused to navigate the AI-mediated world (Leander & Burriss, 2020; South­
mixed reactions from different sectors since its debut in November 2022. worth et al., 2023). Finally, because the new wave of GenAI is impacting
In business, while some businesses have praised its potential to improve human creative endeavors in various domains, including writing,
customer satisfaction and reduce costs, others have raised concerns painting, and music composition, it disrupts traditional paradigms of
about its impact on the quality of customer interactions and the loss of human creativity that were once thought untouchable for AI (Vinchon
jobs (Zarifhonarvar, 2023). In journalism, ChatGPT has been used to et al., 2023). In education, it becomes meaningful to proactively engage

* Corresponding author. 3700 Walnut St, Philadelphia, PA, 19104, USA.

E-mail address: [email protected] (B. Chen).

https://doi.org/10.1016/j.caeai.2023.100184
Received 24 August 2023; Received in revised form 30 October 2023; Accepted 4 November 2023
Available online 11 November 2023
2666-920X/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

students in negotiating their relationships with AI by exploring ways to Many curriculum development efforts have been made to introduce
integrate AI to augment student thinking and creativity (Kasneci et al., AI literacy in K-12. Internationally, several countries have already
2023). started to formally incorporate AI into their primary and secondary
However, AI education in K-12 remains in its infancy, so are tools, education curricula (Cantú-Ortiz et al., 2020; Lee et al., 2021). In the
curricula, and practices for teaching AI literacy (Brummelen et al., 2021; United States, entities spanning federal funding agencies (e.g., the Na­
Leander & Burriss, 2020). According to Long and Magerko (2020), AI tional Science Foundation), corporate leaders (e.g., Google for Educa­
literacy can be broadly defined as “a set of competencies that enables tion), and research institutions have been actively developing or
individuals to critically evaluate AI technologies; communicate and supporting the development of AI curricula in K-12. In China, AI was
collaborate effectively with AI; and use AI as a tool online, at home, and officially introduced into the curriculum in 2018 by the Ministry of
in the workplace” (p. 2). Such definitional work has expanded from Education, followed by the release of the first AI textbooks (Su et al.,
simply treating AI literacy as a computational or technical matter to 2022). Similar efforts were launched in the European Union, Japan,
cover a broad array of fundamental competencies involved in human South Korea, Singapore, and so on. UNESCO has been actively cata­
interaction with AI (Southworth et al., 2023). Ongoing work is seeking lyzing dialogues for more countries to consider AI education (UNESCO,
to develop new curricula for K-12 students to develop AI literacy (Lee 2023). In the meantime, new AI curricular documents are proposed
et al., 2021; Su & Ng, 2023), as well as to create assessment tools to (Touretzky et al., 2019), covering a range of topics including machine
measure AI literacy (Laupichler et al., 2023). But these efforts remain learning (Jiang et al., 2023; Marques et al., 2020; Wan et al., 2020),
preliminary and are yet to reach more classrooms. To advance AI edu­ conversational agents (Van Brummelen, 2019), ethics (Ali et al., 2019),
cation in K-12, we need to advance pedagogical strategies to meaning­ and so forth. The duration of these curricula ranges from hours (Zim­
fully integrate AI in existing subject areas so that AI literacy can be also mermann-Niefield et al., 2020) to a full year (e.g., Sabuncuoglu, 2020).
developed as a by-product of disciplinary learning. To support the teaching of AI literacy in K-12, tools have also been
To contribute to this body of work, this study engages high school created for K-12 students. Notable examples include Teachable Machine
students in not only using GenAI – in particular ChatGPT – in learning (Carney et al., 2020), which features easy-to-use graphical user in­
but also interrogating multiple ways ChatGPT could support their terfaces that allow even young children to explore fundamental AI
collaborative knowledge building. This study represents an effort to look concepts in age-appropriate manners. These tools can also be integrated
beyond simplistic integration of ChatGPT for quick gains (such as with other platforms to achieve various teaching goals. For example,
answering student questions). Rather, it invites students to carefully extending Teachable Machine, the danceOn system (Payne et al., 2021)
configure their relationships with ChatGPT as they work on authentic enables learners to record and creatively program their dance videos,
problems. Through this process, it is conjectured that students can which allows them to explore the intersection between data and dance.
develop a deeper understanding of the strengths and weaknesses of AI, These development efforts cover all K-12 grade levels including early
as well as its potential impact on society. By engaging in this project, childhood education, where games, toys, and physical artifacts are often
high school students are hypothesized to gain valuable experience used to introduce children to AI concepts (Su & Ng, 2023). Physical
working with AI and develop the skills needed to thrive in a world computing toolkits, such as PopBots (Williams et al., 2019), are
increasingly influenced by this technology. commonly used to teach students basic AI concepts.
The remainder of the paper is organized as follows. We first review In terms of pedagogy strategies, teaching AI literacy in K-12
two bodies of literature pertinent to this study, including efforts to teach commonly engage students in hands-on activities and experiential
AI literacy in K-12 and to support learning with AI. After introducing our learning (Su & Ng, 2023; Williams et al., 2023). These approaches are
research questions motivated by the literature, we describe the research often coupled with explainable AI systems that students can manipulate
context, pedagogical design, and data analysis. We then report findings to learn through interacting with such computer systems (Long et al.,
and discuss their implications for both research and practice. 2023). Prior work also involves students to apply AI techniques to
real-world problems, and by doing so students learn both technical skills
2. Relevant work but also critical thinking, problem-solving, and teamwork that are
essential for the field of AI (Southworth et al., 2023). A key goal of
2.1. Teaching AI literacy in K-12 teaching AI literacy is to equip students with the ability to identify
ethical issues and have critical reflections about the societal implications
The importance of teaching AI literacy is increasingly recognized in surrounding the AI (Laupichler et al., 2023). This emphasis is reflected
K-12 education given AI technologies’ increasing intricacy and a great in recent teaching and curriculum design practices, such as embedding
chance they are hidden behind simple user interfaces such as Amazon ethical discussions into students’ hands-on activities (Solyst et al., 2023;
recommendations, Siri, and ChatGPT (Eslami et al., 2016). This is Williams et al., 2023).
especially true for young children, who may have misconceptions about Despite its tremendous growth, the area of introducing AI to K-12
AI concepts and technologies (Su & Ng, 2023), and need to be education is facing a number of challenges. First, in contrast with the
empowered to understand, use, and evaluate AI (Williams et al., 2019; exploding curriculum development efforts contributed by different
Yang, 2019). To many, AI is no longer a purely technical or computa­ players in the space (e.g., researchers, publishers, non-profits, teachers),
tional field (Southworth et al., 2023). AI literacy has greatly expanded empirical evidence of the effectiveness and usefulness of these AI liter­
from teaching technical skills to cover a wide range of topics. For acy programs remains extremely sparse (Brummelen et al., 2021;
instance, Long and Magerko (2020) outline a set of AI competencies that Southworth et al., 2023). Evaluation studies of these programs are
cover fundamental questions about AI, including: What AI is, what AI needed, so are careful research on AI-mediated learning processes.
can do, how AI works, how AI should be used, and how people perceive Relatedly, there is a lack of assessment tools for AI literacy (Brummelen
AI. The identified competencies in their framework are essential for et al., 2021). While AI literacy is closely related to other areas such as
individuals to engage in critical evaluation of AI technologies, effec­ computational thinking, data science, and information literacy, it also
tively communicate and collaborate with AI systems, and utilize AI as a invites the creation of new frameworks that address distinct learning
valuable tool in various contexts such as online platforms, home envi­ objectives regarding AI education (Long & Magerko, 2020). While there
ronments, and workplace. A Delphi study with AI experts has arrived at are attempts to develop survey items to assess AI literacy (Laupichler
a similar list of items that cover how AI works, where AI gets deployed, et al., 2023), this work is far from being mature. Third, educators need
and how biases arise in AI systems (Laupichler et al., 2023). These more and better support for teaching AI literacy either as an indepen­
frameworks provide a rich space for educators to grapple with when dent subject area or across the curriculum (Su & Ng, 2023). Given the
they teach AI literacy in K-12. current “mile-wide, inch-deep” problem with school curriculum (Li,

2
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

2007), seeking to integrate AI literacy in extant subjects might be a more engagement with AI (Markauskaite et al., 2022). Such engagement
fruitful path forward in formal education. More support for teachers to needs to be mindful, using the words of Salomon et al. (1991), in that the
teach AI literacy following interdisciplinary, project-based approaches learner goes through effortful, metacognitive processes of configuring
with suitable technologies is needed (Ng et al., 2023). the human–technology partnership. It needs to be also reflective, so that
students understand how AI-enabled technology can alter their activity
2.2. Integrating AI to support learning in K-12 and can choose how AI is incorporated in learning (Markauskaite et al.,
2022).
Besides explicitly teaching AI literacy in K-12, another related body As AI systems are increasingly deployed in schools, it becomes not
of work deals with using AI to support learning in schools. This work is only useful but imperative to engage students in reflecting on their re­
situated in a long-standing interest in developing and deploying lationships with AI tools because the choice of AI paradigms or the
educational technology in schools to improve learning experiences configuration of AI-human relationships are often reserved to tool pro­
(Huang et al., 2019). Whether it is Skinner’s Teaching Machine in the viders or other capable adults. While the previous section is concerned
early days or modern intelligent tutoring systems, many of these tech­ with what AI competencies are important to cover, a key design
nologies are interested in improving students’ intellectual performance consideration in the second strand is how to meaningfully negotiate
and ability. The same story can be told about using AI to support human-AI relationship in a way that is meaningful and beneficial for
learning. The Artificial Intelligence in Education (AIED) community, for students. This frontier matters for other contexts as well. In domains of
instance, has a long tradition of using AI to support learning and human creativity, generative AI systems engender a new era of “assisted
teaching. Strands of work in the AIED community such as adaptive creativity” in which AI is poised to be a collaborative agent for various
learning, predictive analytics, learner models, and intelligent tutors creative endeavors (Vinchon et al., 2023), building on a long tradition of
leverage various types of AI technologies to promote learning (L. Chen building Creativity Support Tools (Frich et al., 2019). Key to human
et al., 2020). Undergirding these efforts are kaleidoscopic perspectives flourishing in an AI-mediated world is to articulate possible distribution
on human-AI relationships that are believed to boost learning. Explicitly of agency, between human and AI, and between individual and collec­
and intentionally unpacking these hidden beliefs embodied in AI tools tive (Bearman & Ajjawi, 2023; Markauskaite et al., 2022). As students
provides rich opportunities for developing AI literacy in the context of encounter AI tools, it is both productive and imperative to nurture their
learning. reflective practices so that developing AI literacy could to a great extent
First, when deployed to support learning, AI technologies may become a by-product of AI-supported learning.
augment human capability in performing the task, while also enabling
humans to learn from the AI-supported experience. When discussing 2.3. The present study
intelligent technologies for human cognition in the early 1990s, Salo­
mon et al. (1991) distinguish between two concepts: effects with tech­ To explore the use of GenAI for supporting students’ knowledge
nology when humans work in partnership with machines, and effects of work, we situated this study in a classroom that followed a unique
technology shown when humans subsequently work without technol­ pedagogy named Knowledge Building, which aims to parallel knowl­
ogy. They invite the reader to consider: edge creation in organizations (Scardamalia & Bereiter, 2022). In
particular, Knowledge Building aims to foster a knowledge community
“[T]he possible impact of a truly intelligent word processor: On the
where students are positioned as active agents in their learning and
one hand, students might write better while writing with it; on the
build on each other’s ideas to advance collective understanding (Scar­
other hand, writing with such an intelligent word processor might
damalia & Bereiter, 2022). In a Knowledge Building community, stu­
teach students principles about the craft of writing that they could
dents take on a growth mindset for creative work with ideas (B. Chen &
apply widely when writing with only a simple word processor; this
Hong, 2016). They generate theories and ask explanation-seeking
suggests effects of it.” (Salomon et al., 1991, p. 3, p. 3)
questions to solve authentic problems (Zhang et al., 2007). In their di­
This distinction applies to AI tools such as Khan Academy’s newly alogues, students make a range of knowledge-creating dialogue moves
released math tutor named Khanmigo (see https://www.khanacademy. in a non-linear, emergent manner (Bereiter & Scardamalia, 2017, p. 16):
org/khan-labs). This tutor, powered by GPT-4, guides students to learn
through more sophisticated conversations that avoid giving students • Problem definition: Clarifying the essence of the challenge, why it is
direct answers. Khanmigo is seemingly attending to both types of effects: important, and why has it not been addressed already
effects with Khanmigo so that students can successfully solve math • New ideas: Introducing new concepts, distinctions, or analogies and
problems with help from Khanmigo, and effects of Khanmigo so that connecting them to current state of art
students also learn the actual solution from this experience. • Promisingness evaluation: Considering which idea has greater
There are diverse views on human-AI relationships embodied by AI potential
tools—implicitly in most cases—deployed to support learning. In edu­ • Meta-dialogue: Reflecting on the discourse, evaluating its progress,
cation, generative AI can play a range of roles, including personal tutor recognizing individual contributions and collective
that tutors each student and gives immediate feedback, possibility en­ accomplishments
gine that generates alternative ideas, Socratic opponent helps students • Comparison: Looking beyond the immediate sphere to analogous
arguments, and study buddy that helps students reflect on learning problems and solutions, connecting ideas across problem and com­
progress (UNESCO, 2023). In a review, the use cases of AI are charac­ munity boundaries
terized to include three paradigms, namely (a) AI-directed, learner-­ • Critical discourse: Considering the trustworthiness of information on
as-recipient, (b) AI-supported, learner-as-collaborator, and (c) which a particular design decision is based
AI-empowered, learner-as-leader (Ouyang & Jiao, 2021). These para­ • Higher-level ideas: Working collaboratively to develop an idea that
digms are motivated by educational theories and are shaped by factors has application beyond the current problem domain
in specific educational contexts. To a great extent, the AI tools operate
within the boundary of the anticipated paradigm of human-AI rela­ Working in small groups, they engage in progressive inquiry to
tionship. Adopting a relational epistemology, Bearman and Ajjawi improve their theories and synthesize ideas across diverse perspectives
(2023) argue that AI is not defined by its underlying technology but “the to deepen their understanding (Hakkarainen, 2003).
contextually bound relationship between the person and the technol­ While there is a long tradition of using digital technologies and an­
ogy” (p. 4). These arguments move beyond an AI-centered view of AI alytic tools to support Knowledge Building (Chen & Hong, 2016), so far
literacy to consider dispositions and competencies essential for critical as we know no prior work has integrated GenAI in this context.

3
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

Therefore, attempting to integrate GenAI in student knowledge building to absorb content in these materials but identify problems of under­
provides an opportunity to achieve the twin-goals of integrating AI to standing they are interested in and develop ideas to address these
support student creative work and, by doing so, facilitate students’ AI problems (Scardamalia, 2002). The teacher involved students “to foster
literacy. To achieve these goals in the study, we formed a design part­ a Knowledge Building community, where we collaboratively deepen our
nership with a teacher to experiment with concrete strategies of inte­ understanding of these religions and their impact on the contemporary
grating GenAI in knowledge building and posed the following research world” (quote from the course syllabus). Throughout this journey, stu­
questions that are focused on students’ use of ChatGPT and their AI dents participated in discussions, conducted group and individual
literacy: research, created written works, developed presentations, and provided
feedback to their peers.
1. In what ways did students incorporate ChatGPT in their knowledge The use of ChatGPT in the class was to “to support learning and
building? knowledge creation” (quote from the syllabus). The teacher dedicated
2. To what extent was students’ AI literacy enhanced, and how? class time to introducing, discussing, and tinkering with ChatGPT. As
stated in the syllabus: “We will explore different uses of Generative AI,
3. Methods such as ChatGPT, to support the knowledge building process, design-
mode work and learning. We will discuss the properties, capabilities
3.1. Participants and limitations of the large language models that power the latest boom
of AI and explore different prompt engineering techniques that support
One teacher, Mr. F.,1 and 10 high school students (6 boys and 4 girls) learning and research.” Under the rationale and guidance laid out by the
from a World Religions course participated in the study. Students teacher, the process of using ChatGPT could be divided into two phases,
enrolled in the course came from three different schools; they took the when knowledge building in the class was focused on: (1) exploring the
course as an elective selected among a wide range of topics and required problem space, and (2) generating knowledge for the final essay through
for High School graduation at all schools. They were mostly from upper- collaborative discourse. For each phase, we identified a design pattern
middle class families in a major city in South America. Students were which illustrates the integration of AI in knowledge building (see Fig. 1)
native Spanish speakers but the course was conducted in English. All following the knowledge-creating dialogue moves (Bereiter & Scarda­
participants provided informed consent before the study; students were malia, 2017). The following paragraphs offer a detailed explanation for
also provided an option of using the teacher’s ChatGPT account if they the class activities and design patterns for each phase.
did not want to create accounts of their own. Ten other students were Phase 1 (Weeks 5-8): The integration of ChatGPT in this phase could
enrolled in the course but did not consent to the study and were there­ be abstracted to the Design Pattern 1 (DP1) illustrated in Fig. 1 (left side).
fore not included in the study. All students in the class were provided In this pattern, ChatGPT generated an essay with a set of ideas to be
with the same instruction and support regardless of their consent status. critiqued by students and a set of inquiry questions to populate the
The teacher is familiar with the Knowledge Building pedagogy, but problem space for students’ consideration (DP1 - Step 1). Through
students in this course were new to Knowledge Building. Mr. F. also has critical discourse (DP1 - Step 2), students were in charge of evaluating
profound expertise in technology-enhanced learning. To support different ideas and questions (DP1 - Step 3), selecting promising ones
learning and collaboration, he uses a range of digital platforms, such as (DP1 - Step 4), and using them to generate new ideas on Knowledge
Google Classrooms, Miro (a virtual collaborative whiteboard), and Forum (DP1 - Step 5).
Knowledge Forum, and actively explores new technologies. He identifies In Week 5 of the class, Mr. F. formally introduced ChatGPT to the
as an EdTech entrepreneur who teaches part-time, and serves as Chief class. He used ChatGPT to create a sample essay for the students to read
Learning Officer in a startup that actively develops and implements (the prompt provided by Mr. F was shown in Fig. 2, left side). The
learning technologies in the K-12 learning ecosystem. returned essay was fine but too general and lacking in the use of sources,
according to Mr. F. He then used the essay in class as an example of an
3.2. Integrating ChatGPT in knowledge building essay that is well-written but fails to use in-text citations correctly.
Students were asked to grade the essay on a 1–10 scale that reflected
Inspired by an interview with an EdTech professor, Mr. F. was quality of the content and quality of the writing. Students then engaged
convinced that students need to learn how to collaborate with AI. During in critical discourse about the essay and compared ideas in the essay
the first few weeks of the semester, Mr. F. learned from students that with their own ideas. Here, students offered feedback to each other and
they already knew about ChatGPT before the class began. He and the also responded to the question on Miro: “What did ChatGPT’s essay
first author of this paper engaged in frequent email correspondence to contribute to our ideas?.” Finally, they combined ideas from ChatGPT
explore ideas of integrating ChatGPT in his classroom. As public dis­ with their own to identify promising directions of their research, and
cussion and experimentation around ChatGPT was rapidly evolving went forward to generate new ideas.
during the time of the study, the design was constantly informed and In Week 6, students wrapped up their exploration of “What is Reli­
impacted by new information discovered online. For example, while the gion” and started to go deeper in specific topics by identifying promising
concept of prompt engineering was relatively new at the beginning of questions about specific religious traditions. While they generated
the study, a prompt engineering course was quickly released by Deep­ questions in brainstorming activities that followed the Question
Learning.AI, attracting Mr. F. to take it. Throughout the time, Mr. F. was Formulation Technique method (Rothstein et al., 2011), a small team of
leading the emergent design process, with the first author playing a three students volunteered to prompt ChatGPT to generate questions.
supportive role by offering suggestions and technical support. Questions contributed by ChatGPT were posted to a dedicated area on
The central objective of this course was to help students develop the Miro board alongside student-generated questions. Students
religious literacy by investigating major world religions and how they experimented with different prompts and languages (both English and
respond to major contemporary challenges. To support this goal, the Spanish). On Miro, students recorded 59 questions while 29 more were
teacher assigned required readings, which included God Is Not One (by attributed to ChatGPT. In Week 7, students continued to sort and reflect
Stephen Prothero, 2011) as the central text of the course. Other on these questions. Throughout Week 8, they selected questions from
authoritative sources in text, video, and multimedia formats were also the list and continued to build knowledge on Knowledge Forum.
used. Following Knowledge Building pedagogy, students were not asked Phase 2 (Weeks 9-14):The integration of ChatGPT in this phase is
illustrated in the Design Pattern 2 (DP2) in Fig. 1 (right side). To deepen
their knowledge building, students developed specific questions and
1
All names used in the paper are pseudonyms. asked ChatGPT to produce responses to questions (DP2 - Step 1 and 2),

4
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

Fig. 1. Two Patterns of Using ChatGPT

Note. Pattern 1 presents the design for Phase 1. Pattern 2 presents the design for Phase 2. In each pattern, each node is aligned with the knowledge-creating dialogue
moves presented along the vertical axis. Horizontally, each node is aligned with the agent, either student or ChatGPT, who is expected to make the dialogue move.
The number of each node represents the order in which it is anticipated to appear in a sequence of dialogue moves. For example, in Design Pattern 1, the first move
involved having ChatGPT generating “new ideas,” followed by students conducting “critical discourse.”

Fig. 2. Two ChatGPT prompts designed by Mr. F.

5
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

surface different perspectives, and follow-up questions. All students applied these codes to one interview, compared the coding results
were encouraged and supported to use ChatGPT in this process. Under (Cohen’s kappa = .90), addressed disagreements through discussion,
the guidance from Mr. F., students fact-checked the responses, and refined the code book. The finalized code book, as shown in Table 1,
conversed with peers and ChatGPT to generate new ideas (DP2 - Step 3 guided the analysis of the findings through two overarching dimensions:
and 4), reflected on their contributions in the Knowledge Forum (DP2 - (a) “Use AI in Knowledge Building”—pinpointing the integration of AI
Step 4), and worked further on producing higher-level ideas in their into students’ learning experiences, and (b) “AI Literacy”—capturing
essays (DP2 - Step 6). Students also created online posts reflecting on indicators of student understanding of AI. Within these primary di­
their use of ChatGPT. mensions, several sub-codes were further developed including “Utility”,
More specifically, in Week 9, students started to explore deeper “AI-mechanism”, and “AI-relationship”. After establishing a shared un­
questions with ChatGPT. Mr. F. provided specific task instructions to derstanding of the coding scheme, each researcher coded half of the
students, with an articulate workflow that included: (1) converse with remaining data. This iterative process allowed for the integration of
ChatGPT with a steering prompt (see Fig. 2, right side), (2) fact-check predetermined codes with in-depth explanations, ensuring a compre­
with Google or other tools, (3) discuss with peers, (4) write a group hensive understanding of the data. Given the small sample size, this
note on Knowledge Forum with details, and (5) build on other groups’ analysis was less interested in the frequencies of specific codes but more
notes. interested in surfacing student perspectives in related areas.
With the steering prompt, designed by Mr. F., each student posed a To understand the extent to which students’ AI literacy enhanced, we
question to ChatGPT, which then returned a brief answer to the ques­ also investigated the teacher’s notes on Knowledge Forum to triangulate
tion, nuanced viewpoints, and a follow-up question for the student to with the interview results. These notes provided evidence on how the
consider. Students posted their conversation transcripts on Knowledge teacher scaffolded students’ use of AI in varied activities.
Forum, together with a reflection on their experience conversing with
ChatGPT. They found fact-checking very difficult, since ChatGPT sounds
very convincing; they shared that reviewing all the facts would require
multiple Google searches and careful examination of the returned
information.
Table 1
For the rest of the semester, each student continued to build
Coding schemes of using AI in knowledge building and AI literacy.
knowledge about a religion of their choice, with fluid groupings that
Dimension Code Definition
allowed them to move among different groups (of usually 3–4 students)
based on emerging interests, through reading, discussing with peers Use AI in Utility The perceived outcomes and advantages
during class and on Knowledge Forum, and presenting their work to the Knowledge of using AI for learning, collaboration, and
Building creativity.
class for feedback and peer learning. As the final assignment of the class,
Reference: (Long & Magerko, 2020)
students needed to turn in an essay. They were allowed to use ChatGPT Process The ways in which students used AI during
but they were asked to provide ChatGPT transcripts if they used it and class activities, with a specific emphasis
also to properly cite sources. Mr. F. tried to model ethical use of AI by on the procedural aspects of their AI usage
and the specific methods employed by the
clearly identifying AI contributions in a piece of writing while also
students.
providing the prompts and transcripts. The teacher emphasized that the Challenge Challenges students encountered when
essays should be their work, not ChatGPT’s. utilizing AI in the classroom.
Coping strategy Strategies and approaches employed by
3.3. Data sources students to address the challenges they
encountered when utilizing AI in the
classroom.
To answer our research questions, we collected data from multiple AI Literacy AI-mechanism Students’ ability to explain the
sources to investigate the ways students incorporated ChatGPT in functioning of AI systems, including
knowledge building, the nature of their relationship with ChatGPT, and various processes involved in AI and
machine learning.
extent to which their AI literacy was enhanced. Because this study was
Reference: (Laupichler et al., 2023; Long
focused on strategies of integrating ChatGPT and students’ emerging AI & Magerko, 2020)
literacy, we did not collect data about students’ prior knowledge of AI-strength Strengths and advantages of AI systems
ChatGPT or track students’ understanding of the domain area (i.e., perceived by students based on their
world religions). experiences.
Reference: (Laupichler et al., 2023; Long
The primary data source included semi-structured interviews con­
& Magerko, 2020)
ducted with 10 students about their learning experience at the end of the AI-weakness Weaknesses and limitations of AI systems
semester. Students participated in groups of 2 or 4. The first and second reported by the students based on their
authors conducted a total of 4 interviews, each lasting 20 to 40 minutes. experiences.
Reference: (Laupichler et al., 2023)
The interviews were conducted in English, recorded and transcribed
AI-risk Students’ perceived risks that may arise
verbatim. from the use of AI systems such as bias and
Additional secondary data sources were also collected to help us ethical issues.
interpret student perspectives. These secondary data sources included: Reference: (Laupichler et al., 2023; Long
(a) student writing and artifacts from the Miro and Knowledge Forum & Magerko, 2020)
AI-societal The role of AI in society and its potential
generated from collaborative knowledge building during the study; and
implications applications in various domains, both in
(b) teacher planning docs and reflective journals detailing the plan and current and future contexts.
enactment of each week’s class. Reference: (Laupichler et al., 2023; Long
& Magerko, 2020)
AI-relationship Students’ perception of their relationship
3.4. Data analysis
with AI systems that are manifested in
their interaction and connection with AI,
The coding for the interviews was an iterative process. Initially, a their emotional response to AI, and their
code book was developed based on research questions and existing level of trust in AI outputs.
literature while remaining open to emerging patterns. Engaging in Reference: (Bearman & Ajjawi, 2023;
UNESCO, 2023)
descriptive coding (Miles et al., 2019), two researchers independently

6
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

4. Results 4.1.2. Processes of using ChatGPT

The usefulness of ChatGPT was further explicated when students
4.1. Incorporating ChatGPT in knowledge building described concrete processes of integrating ChatGPT in their work.
Students did not merely explore ChatGPT but strategically incorporated
4.1.1. Utility of ChatGPT it at specific junctures of their learning processes, with Mr. F.’s support
Results of student interviews showed students found ChatGPT useful that is described in the Methods section. Students described challenges
for information search, as a serendipity engine, and for accomplishing they encountered during the process and strategies they developed to
mundane learning tasks. cope with them.
ChatGPT, according to a majority of student participants (70%), First, students reported different approaches to integrating ChatGPT
improved their efficiency in searching for information in the process of in writing, with varied levels of cognitive engagement and connection
learning. For example, Luis shared that ChatGPT simplified his infor­ with other digital tools. For example, Valentina frequently sought
mation search process: additional resources and arguments from ChatGPT, while Sophia used
ChatGPT when revising essays. During these processes, students also
“ChatGPT would make searching about those ideas easier. [Unlike]
combined ChatGPT with other tools. For instance, when conversing with
Google, where I would have to shape what I was putting into the
ChatGPT, students turned to Google for fact-checking and additional
search engine, like shape it into a … in a very specific way. In
background research before returning to ChatGPT for further
ChatGPT, I would just put what I was thinking and it basically
conversations.
understood.“2
Despite the wide usage of ChatGPT in writing, students demonstrated
Here, Luis attributed the usefulness of ChatGPT for information varied levels of reliance on the tool. Some fully relied on it to complete
search to its intelligence in easily understanding human queries. Simi­ their homework, while others imposed limits to foster their own critical
larly, other students, including Camila and Javier, recognized thinking. For instance, Camila used ChatGPT as a means to collect in­
ChatGPT’s utility in promptly delivering concise and clear responses. In formation for writing but preferred to synthesize and create her own
comparison with using a search engine, some students appreciated content rather than relying on it entirely. Similarly, Sophia demon­
ChatGPT’s capability in maintaining a chat history, which allows it to strated a critical mindset, carefully evaluating the reliability and quality
better understand students’ emerging needs in an ongoing conversation. of ChatGPT outputs. She mentioned:
As Luis put it, the fact that ChatGPT “remembered the questions above”
“Because I do find myself recognizing that even if ChatGPT or other
led to “in-depth information” that guided his research. Camila also
AI tools can produce great texts, it’s never going to amount to the
valued this feature, and found her interactions with ChatGPT to be very
same quality as when I do it myself. So it’s just a supporting tool. I
“personal.” She felt that “it’s not a place where everyone asks questions;
never rely completely on an AI tool.”
” instead, it was a private space between her and ChatGPT, where the
chat had her “history and everything.” In contrast, Diego reported using ChatGPT sometimes as a shortcut,
In addition to information search, students recognized ChatGPT as a merely copying instructions and pasting ChatGPT outputs into his
valuable tool for supporting knowledge building especially in offering homework. However, he came to realize that this approach hindered his
inspirations, generating new ideas, and promoting collaboration. First, opportunities to learn when the teacher provided guidance during the
students found ChatGPT useful for giving them access to diverse per­ class on how to use ChatGPT meaningfully and responsibly.
spectives, as ChatGPT generated different “facts” and suggested related As their experiences accumulated, students made deliberate efforts
topics beyond the reach of manual research. As Victoria put it, ChatGPT to tame ChatGPT for desirable outputs through prompt engineering. The
provided her with “different ideas” and “different ways to approach a process of prompt engineering, which involves iteratively crafting,
situation,” allowing her to broaden her thinking and generate new ideas. probing, and refining the prompts according to the interaction with
Similarly, Camila commented that ideas provided by ChatGPT inspired LLM, is essential for using ChatGPT (OpenAI, 2022). In this class, the
her to “create new things.” Additionally, about 60% of the participants teacher provided well-crafted and tested prompts in some activities (see
acknowledged using ChatGPT for rapid access to information during Fig. 2), while in others, students interacted with ChatGPT on their own
class discussions so that they could keep up with the conversation and using their own prompt engineering strategies. Many students reflected
make more progress with their peers. that writing effective prompts could be challenging during the process,
However, students had mixed feelings about ChatGPT’s potential in as sometimes ChatGPT did not respond as expected or refused to follow
supporting creativity during knowledge building. On the one hand, specific instructions on word limits or structures. When navigating the
Francisco felt that ChatGPT’s provision of “fact-based” information challenges, Francisco experimented with different words and in­
without opinions limited its contribution to creative thinking. On the structions to elicit the desired response. Alejandro, Miguel, and Luis
other hand, Camila found that the diverse perspectives offered by emphasized the importance of being specific in prompts, using in­
ChatGPT could spark her own creativity. Similarly, Luis believed that structions that could be understood by a “five-year-old kid.” Javier
ChatGPT’s rapid information generation allowed him to efficiently referred to this iterative process of prompt engineering as the “artificial
explore a wider range of ideas, and hereby helped him to be more cre­ intelligence way.” Additionally, they highlighted the need to provide
ative in knowledge building. sufficient contextual information for ChatGPT to understand the
Furthermore, students acknowledged ChatGPT as a useful tool that prompts. Luis, for instance, mentioned starting with general questions to
assisted them in accomplishing mundane learning tasks such as writing establish the context before posing the actual question.
assignments and finishing up homework. For instance, Sophia found In the process of using ChatGPT, students also critically reflected on
ChatGPT beneficial to her for getting started with tasks, as well as cor­ ChatGPT’s limitations and proactively sought support from the teacher
recting minor mistakes in writing assignments. Similarly, Luis utilized when encountering challenges. One major challenge identified by stu­
ChatGPT to create essay outlines, which helped reduce his workload. dents was about the need to verify the accuracy of information provided
Moreover, students also found ChatGPT effective in supporting them to by ChatGPT. To address this issue, while they were incorporating
solve math problems, deliver quick answers to quiz questions, and ChatGPT in learning, many of them mentioned setting boundaries on
generate concise summaries of texts. how AI is used and making conscious decisions on how to incorporate
ChatGPT outputs into their own work. Uncertain about how to properly
cite outputs generated by ChatGPT, Francisco asked for guidance from
2
All quotes in the result section have been edited for clarity, with the original Mr. F., who provided further instructions on creating citations.
meaning remains unchanged.

7
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

4.2. Students’ AI literacy 4.2.2. Strengths and weaknesses of ChatGPT

The interview results indicated that most students recognized
The perceived usefulness of ChatGPT and the processes of using ChatGPT’s strengths in interpreting user prompts, retrieving informa­
ChatGPT were grounded in students’ evolving AI literacy. Student in­ tion efficiently, and offering quick and clear responses. Students
terviews revealed students’ AI literacy in the following important areas. perceived the information provided by ChatGPT to be rich and diverse,
representing different "facts’’ that inspired them to generate new ideas.
4.2.1. Mechanisms of ChatGPT Others appreciated how ChatGPT’s responses were akin to those from a
All students recognized ChatGPT as an AI technology, but their un­ human.
derstanding of how it works varied. The majority had a rudimentary The perceived weaknesses of AI included generating inaccurate or
understanding, considering ChatGPT as a “database of the Internet” that dated information, misinterpreting user input, being limited to certain
detects keywords from human prompts and, in turn, retrieves relevant domains, and operating in a “black box.” Many students highlighted the
information from its “database,” and presents humans with a summary issue of misinformation, with ChatGPT sometimes hallucinating inac­
of the retrieved information. This understanding, while inaccurate, curate information. Students also noted ChatGPT’s lack of access to up-
allowed students to grasp the working mechanism of ChatGPT. For to-date information, given the training data only included sources up to
example, Luis offered the following explanation of ChatGPT’s funda­ September 2021 (when the study was conducted). Additionally, one
mental mechanisms: student expressed concerns about the AI’s “black box” nature that
undermined his trust in information provided by ChatGPT.
“ChatGPT has virtually most of the information that we have avail­
Besides issues with output quality, students also pointed out
able on the internet, so when you would put, like a prompt, or a
ChatGPT’s limited “cognitive” capabilities. To them, ChatGPT is prone
question inside the chat, what I understand is that it like, searches for
to misinterpreting user prompts and providing responses that are either
keywords or, like topics, on everything that it has access to, and it
incomplete or off-topic. For instance, Camila brought up an example
would summarize it into your chat response.”
that ChatGPT did not follow her instruction on the word limit, stating
In this quote, Luis highlighted ChatGPT’s capability in interpreting that “sometimes the directions are not quite what you asked for.” Some
prompts and producing relevant responses by drawing on the massive students also pointed out that ChatGPT’s capabilities were weaker in
dataset. However, he tended to imply that ChatGPT actively searched certain areas such as problem-solving and creativity, especially in do­
the database in real-time, instead of relying on a pre-trained large lan­ mains related to social sciences. For example, Victoria believed that
guage model. Furthermore, he did not delve into the complex deep ChatGPT could brainstorm ideas but could not generate nuanced solu­
learning mechanisms that underpin ChatGPT’s capabilities in inter­ tions to a problem. Similarly, Sophia commented that, “it won’t be able
preting and generating human language. to keep up with our innate creativity and imagination and passion as
This level of understanding, particularly emphasizing the informa­ human beings.” However, it is noteworthy that while they have identi­
tion interpretation and retrieval aspects while neglecting deep learning fied these limitations, students did not have time to further explicate the
mechanisms, appears to be a common trend among teens in this study. origins of these limitations.
However, their views differed on how the database was created. For
instance, Victoria believed “it’s a platform where a lot of articles, data, 4.2.3. Risks and societal implications of ChatGPT
many books, movie[s] have been uploaded,” while Javier elaborated: Students identified risks with using ChatGPT in various areas,
including in school and broader societal applications. First, most of the
“… it has a huge memory and the whole information that other
students were concerned about the potential abuse of AI in school set­
platforms have, such as Google, for every question we ask. There’s
tings, especially for students who, due to lack of access to quality re­
always, like a kind of, I’ll say like a filter, and through that filter,
sources, might not have the opportunities to learn and understand how
specific information from Google or Bing are selected, and then
to use AI properly. For example, Camila stated that “I’ve seen people
they’re paraphrased or formulated as an answer and we obtain that
that used it (ChatGPT) to make something fast and submit the work and
response.”
not even revise it.” Similarly, Sophia mentioned that “I do find it really
These viewpoints reinforced students’ beliefs that ChatGPT queries a dangerous to be a student without knowing the risks and the failures of
database of sources in real-time. Miguel, another student, thought ChatGPT.”
OpenAI developers uploaded files to the database. While Miguel’s The risks go beyond schoolwork. Sophia went further to discuss the
perspective brings developers to the picture, it still implies an active potential impact of AI on her generation that already suffers from a
“uploading” process, rather than the intricate deep learning and rein­ shrinking attention span due to constant exposure to social media:
forcement learning with human feedback mechanisms that undergird
“But again, I feel like it is dangerous … Because, for example, with
ChatGPT (OpenAI, 2022).
TikTok, I feel like I don’t have an attention span anymore. My brain
While the majority of students showed a rudimentary understanding
is constantly consuming this junk and this dopamine. And I cannot
of ChatGPT’s mechanisms, a few delved deeper, demonstrating more
process anything anymore if I’m not overstimulated. So I think it’s a
sophisticated understanding of its intricacies, such as computer algo­
really dangerous combination [of social media and AI]. Like having
rithms powering it and safeguard mechanisms put in place to filter out
this ADHD generation and this tool that makes you everything, I
harmful information. For instance, Valentina explained that “if you put
think that’s really dangerous.”
something that is perceived as bad, ChatGPT is going to tell you, ‘oh, I
can’t answer that because it’s wrong’.” When explaining how ChatGPT Sophia’s responses exemplified a highly critical perspective on the
works, Francisco went beyond describing it as a giant database, risks of AI in both educational settings and society. She repeatedly used
describing ChatGPT’s functioning based on probability: the word “dangerous” to emphasize her concerns. In the same interview
group, Valentina also expressed concerns of humans becoming depen­
“It works based on probability, like you give him a sentence and he
dent on AI as they currently are on the internet. The concerns they
calculates the probability of which word is more probable to be next
brought up were not new or exclusive to AI. They both managed to
in the conversation. And he has, like a database, where he extracts all
connect the use of ChatGPT with other existing societal issues.
that information.”
Besides affecting the wellbeing of individuals, the combination of AI
As shown in the student interviews, student understanding of and social media can also contribute to larger societal issues, such as
ChatGPT’s mechanisms varied in specificity and provided a foundation cyber violence and misinformation. For example, Sophia further
for their critical use of the chatbot. commented:

8
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

“I actually fear that ChatGPT or other AI tools are going to reach the 4.3. How was students’ AI literacy enhanced?
point in which humans use it for violence, for, like cyber violence …
Now I think, can we really start a war? … I feel it’s going to be really The development of students’ AI literacy, demonstrated in the pre­
dangerous and people are really ignorant. And with the combination vious sections, could be attributed to several factors. First, this course
of TikTok and all these sources and the mass media we’re receiving had a positive impact on students’ AI literacy and their use of ChatGPT
constantly and the misinformation we’re constantly consuming, I for learning. It was evident that the teacher’s careful design and inclu­
feel like it’s all going to get out of hand. And that’s why we’re trying sion of ChatGPT as part of the curriculum played a crucial role in
to set boundaries.” enhancing students’ knowledge of AI. For instance, he encouraged stu­
dents to cross-check ChatGPT’s responses using Google. He also
In contrast with acute awareness of these risks, students did not
encouraged students to “play with different types of prompts with more
perceive ChatGPT to be biased. Among all students, six of them dis­
or less context, asking for deep or factual questions, in different lan­
cussed biases in AI. They all thought ChatGPT to be not biased because
guages or with more detail.” Further, he invited students to critically
of the AI’s strict role in responding solely to prompts without any feel­
reflect on ChatGPT’s responses as a way to develop evaluative judgment
ings or opinions. According to the students, ChatGPT’s “neutral” stance
through interactions with AI (Bearman & Ajjawi, 2023).
inherently prevents bias from emerging, as it is designed to remain
Several students, including Francisco, Diego, Javier, and Alejandro,
objective in its responses. One student further added that any perceived
provided further testimonials of the teacher’s guidance being pivotal in
bias would only occur if users deliberately manipulated prompts to lead
their journey with ChatGPT. They learned essential skills, such as proper
the AI towards biased outputs, which she viewed as a misbehavior of the
citation of information and effective prompt engineering, which
user rather than inherent biases within AI. These explanations were
enhanced their overall learning experiences with ChatGPT. For instance,
reasonable and demonstrated their understanding of AI that did not fully
Diego shifted from simply copying and pasting ChatGPT responses to
capture potential biases arising from the training data used for ChatGPT
leveraging it as a learning tool, all thanks to the teacher’s intervention.
(Abramski et al., 2023; Liang et al., 2023; Lucy & Bamman, 2021). For
As a matter of fact, Mr. F intentionally modeled proper use of ChatGPT
example, the model’s training data comprises information gathered
himself and was not reserved from offering direct feedback on students’
from a diverse array of sources and perspectives, which might contain
use of ChatGPT. For example, in one of the Knowledge Forum notes, Mr.
inherent biases reflective of social, cultural, and political viewpoints
F commented on Diego’s work:
present in the data.
Beyond recognizing these risks, students also demonstrated aware­ “The writing is clear and coherent. There are a few minor gram­
ness of other implications of AI on general well-being of society. A matical issues, and, even though you do not include ChatGPT or
majority of students (70%) acknowledged the potential applications of other AIs in the sources, I believe you used some help with your text.
AI in current and future society, particularly in the realms of education I believe it is a mix of your work and the AI’s.”
and the workplace. For instance, some students expressed concerns that
Meanwhile, it was notable that some students had exposure to
relying on ChatGPT might promote laziness and inhibit learning op­
ChatGPT before joining the class through social media or family mem­
portunities and intellectual growth. Miguel commented that while
bers, but the class provided them with sophisticated guidance on using
ChatGPT could improve efficiency at work, it could also “be harmful for
ChatGPT for learning. Camila, for example, mentioned learning about
society like leaving people without a job.”
ChatGPT first from her father, who uses AI on a regular basis in his work.
However, it was through the class that she began exploring AI’s poten­
4.2.4. Relationship with ChatGPT
tial for academic purposes. Miguel, who had prior knowledge of AI from
Students reflected on their relationship with ChatGPT. First, all
social media, acknowledged that the class motivated him to integrate
students perceived ChatGPT to be a valuable tool that greatly supported
ChatGPT in his learning process within the school context in a more
their learning processes by providing information and reducing their
meaningful way.
workload. Notably, some students started to turn to ChatGPT, instead of
Google, as their primary source for information. However, while many
5. Discussion and implications
found ChatGPT helpful, a few students emphasized that they did not
overly depend on it and preferred to set a limit on their usage.
In this study, we involved a class of high-school students to integrate
While ChatGPT was depicted as a valuable tool, students were con­
GenAI in their collaborative knowledge building. In collaboration with
cerned with potential side effects of becoming over-dependent on AI.
the teacher, we derived two design patterns of using ChatGPT to support
Sophia, for example, brought up an essential consideration when dis­
a variety of dialogue moves in knowledge building including problem
cussing the relationship between AI and human intelligence. She
definition, new ideas, and meta-dialogue (Bereiter & Scardamalia,
expressed her concern that relying too heavily on AI-generated ideas
2017). By doing so, we were interested in achieving the twin-goals of
might limit human creativity. To her, ChatGPT should serve as a sup­
both using GenAI to support student creative work and facilitating stu­
portive agent, enhancing human productivity without replacing the need
dents’ AI literacy. This study contributes empirical evidence of K-12
for human imagination and creativity. While strategically integrating AI
students’ use of GenAI in the classroom and the development of AI lit­
into her routines, Sophia stressed the irreplaceable value of human in­
eracy in the process. In this section we discuss main research findings
telligence in certain aspects of learning and problem-solving.
from the study in relation to the literature and propose practical im­
When reflecting on their relationship with AI, students also showed
plications derived from the findings.
varied levels of trust in ChatGPT. Most students admitted that they did
not fully trust AI due to its limitations and ethical considerations, as well
5.1. Discussing main findings
as the potential for misinformation and biases in its responses. However,
Francisco, who demonstrated a more sophisticated understanding of
First, in answering how students incorporated ChatGPT in knowl­
ChatGPT’s innerworkings, surprisingly expressed his complete trust in
edge building, we found that high schoolers used ChatGPT as an infor­
ChatGPT. He commented:
mation generation tool, a serendipity engine, and an assistant for
“Now I trust it, like 100%. After I get something from ChatGPT, I mundane tasks. These usages were to a great extent shaped by the
never check if it is true or not. I just believe it. Maybe when it’s teacher’s design and in line with existing ways of using ChatGPT in
something I know, I am able to identify some things that are not learning contexts (Kasneci et al., 2023; UNESCO, 2023). It was no sur­
correct. But when I don’t know anything about the topic, I will just prise that students turned to ChatGPT to ask domain-specific questions
trust him 100%.” as they inquired about religions. What was worth noting is that they

9
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

used ChatGPT to achieve different goals at different stages of knowledge gathering, and a supportive agent that does not take away human
building instead of solely using it for text generation (Kasneci et al., agency or creativity. While one student would blindly trust ChatGPT’s
2023). outputs, more students did not fully trust ChatGPT due to its various
During the process, students developed strategies for harnessing limitations. Earlier work on AI literacy argues that having accurate
ChatGPT’s strengths while also coping with its weaknesses. Students knowledge of AI’s capabilities in completing different tasks is critical for
leveraged other tools to fact-check ChatGPT outputs, invested time in choosing when to trust AI (Long & Magerko, 2020). In this study, stu­
prompt engineering, and sought advice from the teacher when dents’ understanding of AI mechanisms was limited, as discussed above,
encountering challenges. As pointed out in prior work, while tools like and they showed little awareness of issues of bias in AI. While most
ChatGPT can provide answers quickly, they should never replace stu­ students demonstrated a certain level of criticality toward ChatGPT,
dents’ critical thinking and problem-solving (Dwivedi et al., 2023). By reflection on their personal relationship with ChatGPT could be further
situating the usage of ChatGPT in knowledge building, some students deepened.
believed ChatGPT outputs need to withstand scrutiny before they could
be introduced to their discussion space. Knowledge building provides a 5.2. Implications
meaningful context in which knowledge claims made by either ChatGPT
or students themselves have social and communicative consequences First, the study has practical implications for teaching practice in an
(Knight & Littleton, 2017). It could be argued that while students’ us­ AI-mediated world. To move beyond attention-grabbing news headlines
ages of ChatGPT are not necessarily novel, but the ways in which they such as “ChatGPT killing the student essay,” educators need access to a
incorporated ChatGPT to serve different cognitive functions in a social, rich repertoire of instructional design patterns that tailor AI usage to
discursive environment offers a refreshing take on GenAI’s digital po­ nuanced educational scenarios (Tlili et al., 2023). Volumes of interesting
tential in education (Tlili et al., 2023). In contrast with attempts to use cases of ChatGPT have been crowdsourced (Nerantzi et al., 2023),
provide “personalized” support using ChatGPT (Dwivedi et al., 2023), showing creativity and resourcefulness of educators. This study con­
student use of ChatGPT was personal to them and closely tied to their tributes to this emerging body of principled practical knowledge
communal goal of building knowledge with classmates. (Bereiter, 2014) about using GenAI in education. In particular, when
Results uncovered students’ AI literacy in areas such as AI mecha­ incorporating ChatGPT in knowledge building, we paid special attention
nisms, strengths and weaknesses of AI, risks and societal implications of to the added value of the AI whenever it got used (Luckin et al., 2022).
AI, and human relationship with AI (Long et al., 2023). In particular, The designed patterns presented in the Methods section (see Fig. 1)
students showed rudimentary understanding of ChatGPT, mostly precisely locate ChatGPT intelligence within the knowledge-building
describing it as being powered by a large database of documents while a process and in relation to human intelligence. This approach is akin to
few students showed emerging understanding of the stochastic process efforts in developing “mixed initiative” generative AI interfaces to sup­
of training large language models. Indeed, ChatGPT is powered by GPT port collaborative efforts between humans and generative models
3.5, a large language model trained based on 570 gigabytes of data (Muller et al., 2020). While we do not expect the Knowledge Building
representing 300 billion words (OpenAI, 2022). While students’ pedagogy to be adopted in every classroom, similar efforts could be
knowledge of ChatGPT was still emerging and lacked in specificity, made in other pedagogical contexts (e.g., inquiry learning) to articulate
similar to reports from earlier studies on student understanding of AI in the intended student-AI partnerships.
general (Su et al., 2022), their knowledge served functional roles when As advances in GenAI are forcing us to negotiate boundaries between
they attempted to make sense of ChatGPT. These cases of apparent human and machine intelligences especially in creative domains (Das &
inconsistency revealed by this study are akin to young children’s mental Varshney, 2022; Eshraghian, 2020; Inie et al., 2023), pedagogical stra­
models of the earth in the conceptual change literature (Vosniadou & tegies implemented in the study offers a renewed vision of education for
Brewer, 1992). While perplexing, these results reveal that AI literacy is knowledge creation. In particular, drawing on Knowledge Building’s
complex and would require more than one-time instruction to develop. emphasis on engaging students in continual improvement of ideas
Relatedly, scholars have questioned the extent to which students need to (Scardamalia & Bereiter, 2022), this study shows the potential of
grapple with “black-boxed” AI models, and whether it would be wiser to leveraging GenAI to further augment students’ creativity and epistemic
involve students to work with opaque, partial, and ambiguous situations agency. This particular use of ChatGPT is in contrast with using the
when dealing with these AI models (Bearman & Ajjawi, 2023). Our chatbot to support knowledge acquisition (Kasneci et al., 2023). Re­
finding of students’ limited knowledge of AI mechanisms provides a searchers have advanced three distinct metaphors of learning, including
glimpse into the reality. Despite limited understanding of AI mecha­ learning as knowledge acquisition, social participation, and knowledge
nisms, students demonstrated appreciation of ChatGPT’s strengths in creation (Paavola et al., 2004). As GenAI pushes society to redefine
understanding user prompts and generating relevant responses, as well domains of human work, to realize the potential of GenAI in education
as its limitations in providing accurate, up-to-date information, being (Tlili et al., 2023), we need to be mindful of the visions of learning
interpretable, and strictly following user inputs. The understanding of undergirding each use case of GenAI. This study illuminates the poten­
these limitations, which are noted in both white papers from OpenAI tial of supporting human-AI co-creativity (Lin et al., 2023), motivating a
(2023) and independent research (Alkaissi & McFarlane, 2023), shift from traditional knowledge acquisition to more dynamic learning
emerged from students’ real experience of using ChatGPT. In response to experiences. As GenAI gets introduced to education, we should seek
the need for greater attention to AI in education (Southworth et al., opportunities to transform traditional school practices and in­
2023), besides teaching fundamental AI mechanisms (e.g., Jiang et al., frastructures, as well as views of learning they embody, into new ones
2023), this study suggests the prospect of instrumenting rich opportu­ that can better serve students in an AI-mediated world.
nities for students to develop lived experiences of using GenAI in This study also suggests the need for creating opportunities for stu­
meaningful scenarios. dents to build understanding of GenAI and reflect on their personal
Besides understanding of AI mechanisms and these strengths and relationship with GenAI. In the study, students demonstrated mindful
weaknesses, students demonstrated sharpened insights into potential engagement with ChatGPT (Salomon et al., 1991), as they went through
risks and implications of GenAI for a society in which social media is effortful, metacognitive processes of interacting with ChatGPT for
already complicating people’s personal habits, attention span, and in­ knowledge building. While the teacher has provided careful scaffolds,
formation ecology. Many of these risks pointed out by students are students’ understanding of how AI works and how biases may arise
aligned with current dialogues in the literature (Dwivedi et al., 2023; remained lacking even though they showed considerable criticality to­
Tlili et al., 2023). When reflecting on their relationships with GenAI, ward using ChatGPT for learning. When introducing GenAI to the
students perceived ChatGPT as a valuable assistant for information classroom, teachers should create dedicated space for students to

10
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

develop their AI literacy in these areas. Importantly, robust professional Bereiter, C. (2014). Principled practical knowledge: Not a bridge but a ladder. The
Journal of the Learning Sciences, 23(1), 4–17. https://doi.org/10.1080/
learning opportunities need to be created for teachers to take on this
10508406.2013.812533
challenge while leveraging their existing repertoire of expertise. Bereiter, C., & Scardamalia, M. (2017). “Good Moves” in knowledge-creating dialogue.
Qwerty-Open and Interdisciplinary Journal of Technology, Culture and Education, 11(2),
12–26.
6. Conclusions Brummelen, J. V., Heng, T., & Tabunshchyk, V. (2021). Teaching tech to talk: K-12
conversational artificial intelligence literacy curriculum and development tools.
As a future of partnering humans with GenAI becomes inevitable, Proceedings of the AAAI Conference on Artificial Intelligence, 35(17). https://doi.org/
10.1609/aaai.v35i17.17844. Article 17.
supporting mindful engagement in the partnership is crucial. Such Cantú-Ortiz, F. J., Galeano Sánchez, N., Garrido, L., Terashima-Marin, H., & Brena, R. F.
mindful engagement needs to be reflective, effortful, and metacognitive, (2020). An artificial intelligence educational strategy for the digital transformation.
and demands new competencies of humans as well as guardrails to International Journal on Interactive Design and Manufacturing, 14(4), 1195–1209.
https://doi.org/10.1007/s12008-020-00702-8
minimize undesired consequences. While education could benefit from Carney, M., Webster, B., Alvarado, I., Phillips, K., Howell, N., Griffith, J., Jongejan, J.,
intelligent technologies such as GenAI, we also believe it is important to Pitaru, A., & Chen, A. (2020). Teachable machine: Approachable web-based tool for
create opportunities for students to interact with GenAI mindfully in exploring machine learning classification. In Extended abstracts of the 2020 CHI
conference on human factors in computing systems (Vols. 1–8). https://doi.org/
authentic manners so that they could develop AI literacy while forging 10.1145/3334480.3382839
partnerships with AI for learning. This study makes an initial step to Chen, L., Chen, P., & Lin, Z. (2020). Artificial intelligence in education: A review. IEEE
integrate ChatGPT and GenAI in students’ collaborative knowledge Access, 8, 75264–75278. https://doi.org/10.1109/ACCESS.2020.2988510
Chen, B., & Hong, H.-Y. (2016). Schools as knowledge-building organizations: Thirty
building. Following designs created by the teacher and researchers,
years of design research. Educational Psychologist, 51(2), 266–288. https://doi.org/
students used ChatGPT to support specific knowledge-building moves. 10.1080/00461520.2016.1175306
Findings from the study showed students’ capabilities in leveraging Das, P., & Varshney, L. R. (2022). Explaining artificial intelligence generation and
ChatGPT for knowledge building despite their rudimentary and some­ creativity: Human interpretability for novel ideas and artifacts. IEEE Signal Processing
Magazine, 39(4), 85–95. https://doi.org/10.1109/MSP.2022.3141365
times inconsistent understandings of ChatGPT. The study offers a fresh Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K.,
vision of leveraging GenAI to augment students’ creativity while push­ Baabdullah, A. M., Koohang, A., Raghavan, V., Ahuja, M., Albanna, H.,
ing us to ask several resounding questions: How does students’ mindful Albashrawi, M. A., Al-Busaidi, A. S., Balakrishnan, J., Barlette, Y., Basu, S., Bose, I.,
Brooks, L., Buhalis, D., … Wright, R. (2023). Opinion paper: “So what if ChatGPT
engagement with GenAI look like in K-12? What is needed—from stu­ wrote it?” Multidisciplinary perspectives on opportunities, challenges and
dents, teachers, and others—to nurture such mindful engagement? How implications of generative conversational AI for research, practice and policy.
may such mindful engagement impact student learning? The scope of International Journal of Information Management, 71, Article 102642. https://doi.org/
10.1016/j.ijinfomgt.2023.102642
this study was necessarily limited to strategies of using ChatGPT for Eshraghian, J. K. (2020). Human ownership of artificial creativity. Nature Machine
learning and did not examine the impact of such student-AI partnership Intelligence, 2(3). https://doi.org/10.1038/s42256-020-0161-x. Article 3.
on learning. Much more work is needed to answer these questions, as Eslami, M., Karahalios, K., Sandvig, C., Vaccaro, K., Rickman, A., Hamilton, K., &
Kirlik, A. (2016). First I “like” it, then I hide it: Folk theories of social feeds. In
well as to investigate the impact of student-AI partnerships on various Proceedings of the 2016 CHI conference on human factors in computing systems (pp.
learning outcomes in diverse settings. 2371–2382). https://doi.org/10.1145/2858036.2858494
Frich, J., MacDonald Vermeulen, L., Remy, C., Biskjaer, M. M., & Dalsgaard, P. (2019).
Mapping the landscape of creativity support tools in HCI. Proceedings of the 2019 CHI
Statements on open data and ethics Conference on Human Factors in Computing Systems, 1–18.
Huang, R., Spector, J. M., & Yang, J. (2019). Introduction to educational technology. In
The study has been approved by the University of Pennsylvania. R. Huang, J. M. Spector, & J. Yang (Eds.), Educational technology (pp. 3–31). Springer
Singapore. https://doi.org/10.1007/978-981-13-6643-7_1.
Research data comprised student interviews and the approved IRB Inie, N., Falk, J., & Tanimoto, S. (2023). Designing participatory AI: Creative
protocol does not permit further sharing beyond the authors. professionals’ worries and expectations about generative AI. In Extended abstracts of
the 2023 CHI conference on human factors in computing systems (Vols. 1–8). https://
doi.org/10.1145/3544549.3585657
Declaration of generative AI and AI-assisted technologies in the Jiang, S., Tang, H., Tatar, C., Rosé, C. P., & Chao, J. (2023). High school students’ data
writing process modeling practices and processes: From modeling unstructured data to evaluating
automated decisions. Learning, Media and Technology, 0(0), 1–19. https://doi.org/
10.1080/17439884.2023.2189735
None. Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F.,
Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G.,
Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., …
Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large
Declaration of competing interest language models for education. Learning and Individual Differences, 103, Article
102274. https://doi.org/10.1016/j.lindif.2023.102274
Knight, S., & Littleton, K. (2017). Socialising epistemic cognition. Educational Research
None.
Review, 21, 17–32. https://doi.org/10.1016/j.edurev.2017.02.003
Laupichler, M. C., Aster, A., & Raupach, T. (2023). Delphi study for the development and
Appendix A. Supplementary data preliminary validation of an item set for the assessment of non-experts’ AI literacy.
Computers & Education: Artificial Intelligence, 4, Article 100126. https://doi.org/
10.1016/j.caeai.2023.100126
Supplementary data to this article can be found online at https://doi. Leander, K. M., & Burriss, S. K. (2020). Critical literacy for a posthuman world: When
org/10.1016/j.caeai.2023.100184. people read, and become, with machines. British Journal of Educational Technology,
51(4), 1262–1276. https://doi.org/10.1111/bjet.12924
Lee, I., Ali, S., Zhang, H., DiPaola, D., & Breazeal, C. (2021). Developing middle school
References students’ AI literacy. In Proceedings of the 52nd ACM technical symposium on computer
science education, 191. https://doi.org/10.1145/3408877.3432513. –197.
Li, J. (2007). Bridging across the mile-wide and mile-deep chasm. Standards in Education,
Abramski, K., Citraro, S., Lombardi, L., Rossetti, G., & Stella, M. (2023). Cognitive
7, 33.
network science reveals bias in GPT-3, GPT-3.5 turbo, and GPT-4 mirroring math
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased
anxiety in high-school students. Big Data and Cognitive Computing, 7(3). https://doi.
against non-native English writers. Patterns, 4(7), Article 100779. https://doi.org/
org/10.3390/bdcc7030124. Article 3.
10.1016/j.patter.2023.100779
Ali, S., Payne, B. H., Williams, R., Park, H. W., & Breazeal, C. (2019). Constructionism,
Lin, Z., Ehsan, U., Agarwal, R., Dani, S., Vashishth, V., & Riedl, M. (2023). Beyond
ethics, and creativity: Developing primary and middle school artificial intelligence
prompts: Exploring the design space of mixed-initiative Co-creativity systems.
education. International Workshop on Education in Artificial Intelligence K-12
International Conference on Computational Creativity.
(Eduai’19), 2, 1–4.
Long, D., & Magerko, B. (2020). What is AI literacy? Competencies and design
Alkaissi, H., & McFarlane, S. I. (2023). Artificial hallucinations in ChatGPT: Implications
considerations. In Proceedings of the 2020 CHI conference on human factors in
in scientific writing. Cureus. https://doi.org/10.7759/cureus.35179
computing systems (pp. 1–16). https://doi.org/10.1145/3313831.3376727
Bearman, M., & Ajjawi, R. (2023). Learning to work with the black box: Pedagogy for a
Long, D., Roberts, J., Magerko, B., Holstein, K., DiPaola, D., & Martin, F. (2023). AI
world with artificial intelligence. British Journal of Educational Technology. https://
literacy: Finding common threads between education, design, policy, and
doi.org/10.1111/bjet.13337. n/a(n/a).

11
B. Chen et al. Computers and Education: Artificial Intelligence 5 (2023) 100184

explainability. In Extended abstracts of the 2023 CHI conference on human factors in Solyst, J., Xie, S., Yang, E., Stewart, A. E., Eslami, M., Hammer, J., & Ogan, A. (2023). “I
computing systems (pp. 1–6). https://doi.org/10.1145/3544549.3573808 would like to design”: Black girls analyzing and ideating fair and accountable AI. In
Luckin, R., Cukurova, M., Kent, C., & du Boulay, B. (2022). Empowering educators to be Proceedings of the 2023 CHI conference on human factors in computing systems (pp.
AI-ready. Computers & Education: Artificial Intelligence, 3, Article 100076. https://doi. 1–14). https://doi.org/10.1145/3544548.3581378
org/10.1016/j.caeai.2022.100076 Southworth, J., Migliaccio, K., Glover, J., Glover, J., Reed, D., McCarty, C.,
Lucy, L., & Bamman, D. (2021). Gender and representation bias in GPT-3 generated Brendemuhl, J., & Thomas, A. (2023). Developing a model for AI across the
stories. Proceedings of the Third Workshop on Narrative Understanding, 48–55. https:// curriculum: Transforming the higher education landscape via innovation in AI
doi.org/10.18653/v1/2021.nuse-1.5 literacy. Computers & Education: Artificial Intelligence, 4, Article 100127. https://doi.
Markauskaite, L., Marrone, R., Poquet, O., Knight, S., Martinez-Maldonado, R., org/10.1016/j.caeai.2023.100127
Howard, S., Tondeur, J., De Laat, M., Buckingham Shum, S., Gašević, D., & Su, J., & Ng, D. T. K. (2023). Artificial intelligence (AI) literacy in early childhood
Siemens, G. (2022). Rethinking the entwinement between artificial intelligence and education: The challenges and opportunities. Computers & Education: Artificial
human learning: What capabilities do learners need for a world with AI? Computers Intelligence. , Article 100124. https://doi.org/10.1016/j.caeai.2023.100124
& Education: Artificial Intelligence, 3, Article 100056. https://doi.org/10.1016/j. Su, J., Zhong, Y., & Ng, D. T. K. (2022). A meta-review of literature on educational
caeai.2022.100056 approaches for teaching AI at the K-12 levels in the Asia-Pacific region. Computers &
Marques, L. S., Gresse von Wangenheim, C., & Hauck, J. C. (2020). Teaching machine Education: Artificial Intelligence, 3, Article 100065. https://doi.org/10.1016/j.
learning in school: A systematic mapping of the state of the art. Informatics in caeai.2022.100065
Education, 19(2), 283–321. Tlili, A., Shehata, B., Adarkwah, M. A., Bozkurt, A., Hickey, D. T., Huang, R., &
Miles, M. B., Huberman, A. M., & Saldana, J. (2019). Qualitative data analysis: A methods Agyemang, B. (2023). What if the devil is my guardian angel: ChatGPT as a case
sourcebook (4th ed.). SAGE Publication. study of using chatbots in education. Smart Learning Environments, 10(1), 15. https://
Muller, M., Weisz, J. D., & Geyer, W. (2020). Mixed initiative generative AI interfaces: An doi.org/10.1186/s40561-023-00237-x
analytic framework for generative AI applications. The Future of Co-Creative Systems A Touretzky, D., Gardner-McCune, C., Martin, F., & Seehorn, D. (2019). Envisioning AI for
Workshop on Human-Computer Co-Creativity. K-12: What should every child know about AI? Proceedings of the AAAI Conference on
Nerantzi, C., Abegglen, S., & Karatsiori, M. (2023). 101 creative ideas to use AI in Artificial Intelligence, 33(1). https://doi.org/10.1609/aaai.v33i01.33019795. Article
education, A crowdsourced collection (2023 1.0). A.A. Martínez (Ed.). Zenodo. https:// 01.
doi.org/10.5281/zenodo.8072950 UNESCO. (2023). ChatGPT and artificial intelligence in higher education: A quick start guide.
Ng, D. T. K., Lee, M., Tan, R. J. Y., Hu, X., Downie, J. S., & Chu, S. K. W. (2023). A review UNESCO. https://www.iesalc.unesco.org/wp-content/uploads/2023/04/ChatGPT
of AI teaching and learning from 2000 to 2020. Education and Information -and-Artificial-Intelligence-in-higher-education-Quick-Start-guide_EN_FINAL.pdf.
Technologies, 28(7), 8445–8501. https://doi.org/10.1007/s10639-022-11491-w Van Brummelen, J. (2019). Conversational agents to democratize artificial intelligence.
OpenAI. (2022). Introducing ChatGPT. https://openai.com/blog/chatgpt. 2019 IEEE Symposium on visual Languages and human-centric computing (VL/HCC),
OpenAI. (2023). GPT-4 System Card. OpenAI https://cdn.openai.com/papers/gpt-4-syst 239–240 https://doi.org/10.1109/VLHCC.2019.8818805.
em-card.pdf. Vinchon, F., Lubart, T., Bartolotta, S., Gironnay, V., Botella, M., Bourgeois, S.,
Ouyang, F., & Jiao, P. (2021). Artificial intelligence in education: The three paradigms. Burkhardt, J.-M., Bonnardel, N., Corazza, G. E., Glaveanu, V., Hanson, M. H.,
Computers & Education: Artificial Intelligence, 2, Article 100020. https://doi.org/ Icevic, Z., Karwowski, M., Kaufman, J. C., Okada, T., Reiter-Palmon, R., &
10.1016/j.caeai.2021.100020 Gaggioli, A. (2023). Artificial intelligence & creativity: A manifesto for collaboration.
Paavola, S., Lipponen, L., & Hakkarainen, K. (2004). Models of innovative knowledge https://doi.org/10.31234/osf.io/ukqc9. PsyArXiv.
communities and three metaphors of learning. Review of Educational Research, 74(4), Vosniadou, S., & Brewer, W. F. (1992). Mental models of the earth: A study of conceptual
557–576. https://doi.org/10.3102/00346543074004557 change in childhood. Cognitive Psychology, 24(4), 535–585. https://doi.org/10.1016/
Pavlik, J. V. (2023). Collaborating with ChatGPT: Considering the implications of generative 0010-0285(92)90018-W
artificial intelligence for journalism and media education. Journalism & Mass Wan, X., Zhou, X., Ye, Z., Mortensen, C. K., & Bai, Z. (2020). SmileyCluster: Supporting
Communication Educator, Article 10776958221149576. https://doi.org/10.1177/ accessible machine learning in K-12 scientific discovery. Proceedings of the Interaction
10776958221149577 Design and Children Conference, 23–35. https://doi.org/10.1145/3392063.3394440
Payne, W. C., Bergner, Y., West, M. E., Charp, C., Shapiro, R. B., Szafir, D. A., Williams, R., Ali, S., Devasia, N., DiPaola, D., Hong, J., Kaputsos, S. P., Jordan, B., &
Taylor, E. V., & DesPortes, K. (2021). danceOn: Culturally responsive creative Breazeal, C. (2023). AI + ethics curricula for middle school youth: Lessons learned
computing. In Proceedings of the 2021 CHI conference on human factors in computing from three project-based curricula. International Journal of Artificial Intelligence in
systems (pp. 1–16). https://doi.org/10.1145/3411764.3445149 Education, 33(2), 325–383. https://doi.org/10.1007/s40593-022-00298-y
Russell, S., & Norvig, P. (1995). A modern, agent-oriented approach to introductory Williams, R., Park, H. W., & Breazeal, C. (2019). A is for artificial intelligence: The
artificial intelligence. ACM SIGART Bulletin, 6(2), 24–26. https://doi.org/10.1145/ impact of artificial intelligence activities on young children’s perceptions of robots.
201977.201989 In Proceedings of the 2019 CHI conference on human factors in computing systems (Vols.
Sabuncuoglu, A. (2020). Designing one year curriculum to teach artificial intelligence for 1–11). https://doi.org/10.1145/3290605.3300677
middle school. In Proceedings of the 2020 ACM conference on innovation and technology Yang, X. (2019). Accelerated move for AI education in China. ECNU Review of Education,
in computer science education (pp. 96–102). https://doi.org/10.1145/ 2(3), 347–352. https://doi.org/10.1177/2096531119878590
3341525.3387364 Zarifhonarvar, A. (2023). Economics of ChatGPT: A labor market View on the occupational
Salomon, G., Perkins, D. N., & Globerson, T. (1991). Partners in cognition: Extending Impact of artificial intelligence (ssrn scholarly paper 4350925). https://doi.org/
human intelligence with intelligent technologies. Educational Researcher, 20(3), 2–9. 10.2139/ssrn.4350925
https://doi.org/10.3102/0013189X020003002 Zhang, J., Scardamalia, M., Lamon, M., Messina, R., & Reeve, R. (2007). Socio-cognitive
Scardamalia, M. (2002). Collective cognitive responsibility for the advancement of dynamics of knowledge building in the work of 9- and 10-year-olds. Educational
knowledge. In B. Smith (Ed.), Liberal education in a knowledge society (pp. 67–98). Technology Research & Development, 55(2), 117–145. https://doi.org/10.1007/
Open Court. s11423-006-9019-0
Scardamalia, M., & Bereiter, C. (2022). Knowledge building and knowledge creation. In Zimmermann-Niefield, A., Polson, S., Moreno, C., & Shapiro, R. B. (2020). Youth making
R. K. Sawyer (Ed.), The cambridge handbook of the learning sciences (3rd ed., pp. machine learning models for gesture-controlled interactive media. Proceedings of the
385–405). Cambridge University Press. https://doi.org/10.1017/ Interaction Design and Children Conference, 63–74. https://doi.org/10.1145/
9781108888295.024. 3392063.3394438

12