Computers and Education: Artificial Intelligence 4 (2023) 100124

Contents lists available at ScienceDirect

Computers and Education: Artificial Intelligence

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

Artificial Intelligence (AI) Literacy in Early Childhood Education: The

Challenges and Opportunities
Jiahong Su *, Davy Tsz Kit Ng, Samuel Kai Wah Chu
Faculty of Education, The University of Hong Kong, Hong Kong, China

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

Keywords: Nowadays, Artificial Intelligence (AI) literacy has become an emerging topic in digital literacy education
AI research. However, it is still under-explored in early childhood education (ECE) since the AI curriculum for
AI literacy young children has just been designed in recent years. A scoping review was conducted to examine the thematic
AI education
and content analysis of 16 empirical papers from 2016 to 2022. This scoping reviews evaluate, synthesize, and
Learning and teaching AI
Early childhood education
display 16 studies on AI literacy in early childhood education, including curriculum design, AI tools, pedagogical
Challenges and opportunities approaches, research designs, assessment methods, and findings. The discussion of the AI literacy implementa­
tion in ECE contributes to providing references for educators and researchers to design interventions to engage
young children in AI learning. Further, we identified a set of challenges and opportunities of AI literacy. Several
challenges included (1) lack of teachers’ AI knowledge, skills, and confidence; (2) lack of curriculum design; and
(3) lack of teaching guidelines. Although educators meet challenges at the beginning stage of developing AI
instructional design for young children, AI learning could bring learning opportunities and foster young chil­
dren’s AI literacy in terms of AI concepts, practices and perspectives. We foresee that there will be a growing
number of age-appropriate curriculum and tools for the ECE level. At the end, we also make some recommen­
dations for future researchers and educators to improve AI literacy research and learning design in early
childhood education.

1. Introduction behind these tools and they may have misconceptions about the tech­
nologies. Moreover, AI may threaten their safety when it gives wrong
In 1956, artificial intelligence (AI) was defined as “the science and and misleading information or suggestions (Gaube et al., 2021).
engineering of creating intelligent machines” (McCarthy, 2007). AI re­ Therefore, it is necessary to develop young children’s AI literacy,
fers to the science and engineering of making intelligent machines that especially its limitations, ethical concerns and basic understanding
solve different kinds of problems via natural language processing, neural about the technology (Kong et al., 2021, 2022; Long & Magerko, 2020;
network and machine learning (Mondal, 2020). It is transforming every Ng et al., 2021a, b).
walks of life such as medicine, psychology, science and public policy (Xu Recent researchers proposed the term “AI literacy” to put forth the
et al., 2021). In educational fields, AI can help teachers to predict stu­ importance of adding AI to the 21st century digital literacy skills for
dents’ learning status and performance, recommend learning resources, everyone, including young children (Ng et al., 2021a, b). Learning
and automate assessments to improve students’ learning experience via programs and activities (e.g., Jibo robot, Anki’s Cozmo robot) emerged
intelligent agent systems, chatbots, and recommendation systems (Liang to focus on how to foster young children’s AI understandings and atti­
et al., 2021; Mousavinasab et al., 2021; Su et al., 2022; Zawacki-Richter tudes (Druga et al., 2021). AI literacy has become an essential literacy
et al., 2019; Zheng et al., 2021). skill that is required for everyone (including young children) to know
Young children are growing up with these AI applications, yet little and use AI as a tool to live, learn, and work in our digital world, and it
attention is paid to the important of developing AI literacy and its should be taught in grades K-12 (Burgsteiner et al., 2016; Kandlhofer
related impacts among them. They use AI products such as chatbots and et al., 2016; Ng et al., 2021a, b; Steinbauer et al., 2021). On top of
recommendation tools to facilitate their everyday lives and study. merely becoming end users of AI tools, AI literacy is a set of compe­
Children may not know how to use AI, and the basic working principles tencies that enables people to critically evaluate, communicate and

* Corresponding author. The University of Hong Kong, Room 219, Runme Shaw Building, Pokfulam Road, Hong Kong, China.
E-mail address: maggiesu@connect.hku.hk (J. Su).

https://doi.org/10.1016/j.caeai.2023.100124
Received 11 October 2022; Received in revised form 30 November 2022; Accepted 9 January 2023
Available online 11 January 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/).
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

collaborate effectively with AI (Druga et al., 2021). Scholars started to 2.1. Artificial intelligence
suggest different models to conceptualize the term AI literacy (Ng et al.,
2021b; Touretzky et al., 2019). For example, Touretzky et al. (2019) Recent years, a growing number of researchers have started to
proposed the use of five big ideas about AI: (1) Computers perceive the discuss how to teach and learn AI from K-12 to higher education.
world using sensors, (2) Agents maintain models/representations of the Although the AI-related ECE studies is still in its infancy, researchers
world and use them for reasoning, (3) Making agents interact comfort­ have started to discover how AI applications is used to facilitate
ably with humans is a substantial challenge for AI developers, (4) kindergarten teachers’ administration and students’ learning through
Computers can learn from data, and (5) AI applications can impact so­ intelligent tutoring systems for special education, chatbots for language
ciety in both positive and negative ways. Ng et al. (2021b) further education, and robotic kits in Computer Science education (e.g., Chen
designed a framework of AI concepts, practices and perspectives that et al., 2020; Su & Yang, 2022). These AI applications facilitate students
interplay between CT and AI that students can learn machine learning in computer-supported collaborative learning, teaching automation and
knowledge, model training skills, collaboration and communication evaluation, detecting learners’ emotions, and recommend useful mate­
skills. rials for students. Several facts were uncovered in Chen et al. (2020)’s
In the past, AI learning began in university computer science edu­ paper about how to use AI technologies in the educational industry.
cation, which required advanced programming skills that were inap­ Although a growing interest in and the impact of research on AI in ed­
propriate for children decades ago (Ng et al., 2021a). Recently, the ucation are identified, further effort is necessary to inform how to
development of more age-appropriate software has enabled young integrate advanced AI techniques and deep learning technologies into
learners to extend their possibilities to learn and explore AI. The ma­ educational settings. Also, there exists a scarce number of studies that
jority of current AI literacy research focuses on secondary or higher drive the use of AI technologies into educational theories. This is
education (Ng & Chu, 2021; Kong et al., 2021; Eguchi et al., 2021; Su consistent with another review conducted by Hwang et al. (2021), which
et al., 2022). For example, non-computer science undergraduates and categorised AI applications into four general roles, including intelligent
secondary students started to develop AI concepts and ethical awareness tutor, tutee, learning tool/partner, and policy-making advisor.
to empower them to become educated digital citizens (Kong et al., 2023; In early childhood education, educators tried to explore the use of
Ng & Chu, 2021). However, these courses do not focus on complex AIED technologies to facilitate their work and enhance students’
computer science concepts or mathematical formulas. Instead, the learning. For example, Jin (2019) explored the potential of AI applica­
courses enabled students to develop basic understanding of AI concepts, tions in four examples: AI evaluation of children, AI teaching system, AI
literacy and confidence of using AI (Kong et al., 2022; Ng et al., 2022a). educational robot, and AI virtual reality teaching, and further proposed
In early childhood education, AI-powered toys are made to provide the use of AI in family education and enhanced parental knowledge. Lin
students with a playful experience to learn and interact with the robots et al. (2020) interacted with a dialogue system that enabled students to
and kits, and teach coding skills. With more well-designed AI toys and interact with the chatbots to enhance language learning and visualiza­
services, young children could develop their AI literacy even at a tion training. Nan (2020) used an AI teaching system to motivate stu­
kindergarten level. They experience AI-driven robotic toys and services dents’ learning in a collaborative AI-assisted environment and stimulate
(e.g., PopBots, Quickdraw) (Williams, 2018; Williams et al., 2019a,b) to children’s interest in learning. These studies provide ample evidence to
explore AI related concepts (e.g., knowledge-based systems, supervised show the effectiveness of using AIED technologies at kindergarten level.
machine learning, generative AI). Although young children may not Furthermore, it is essential to equip children with digital skills and
know and understand the knowledge behind, they can explore and mindsets to get them ready for future studies and facilitate their
appreciate these AI technologies, and foster their digital literacy in their everyday living.
everyday lives.
Some may challenge whether kindergarteners and junior primary 2.2. AI literacy
students are too young to explore and learn AI knowledge (Su et al.,
2022). However, previous studies have brought AI tools into early The term “AI literacy” was first coined by Burgsteiner et al. (2016)
childhood education (ECE) classrooms and shown their promising ef­ and Kandlhofer et al. (2016) who describes the competencies to un­
fects (e.g., Lin et al., 2020; Tseng et al., 2021; Williams et al., 2019). In derstand the basic knowledge and concepts about AI. On top of this,
fact, early AI literacy is important for young children to improve many Long and Magerko (2020) defined it as a set of competencies that en­
aspects of child development, such as theory of mind skills, creative ables individuals to critically evaluate, communicate and collaborate
inquiry, emotional inquiry and collaborative inquiry (Kewalramani effectively with AI; and use AI as a tool online, at home, and in the
et al., 2021; Su & Yang, 2022). For example, Kewalramani et al. (2021) workplace. Further, Ng et al. (2021a, b) added AI to every student’s
investigated the use of interactive AI to cultivate inquiry literacy in early twenty-first century digital literacy in work settings and everyday life,
childhood education settings. However, only a few research efforts had and proposed it as a fundamental skill for everyone, not just for com­
been done on how AI literacy for children aged 3–8 years in ECE level puter scientists. He widely incorporated the model of Bloom Taxonomy,
relative to other age groups (e.g, primary 3-6, secondary education) to Technological Pedagogical Content Knowledge (TPACK), and AI con­
support the potential of AI literacy education for ECE settings. So far, cepts, practices, and perspectives into the instructional design of AI
only one review study has tried to discuss using AI in ECE settings (Su & literacy education.
Yang, 2022). However, this review did not discuss the mapping of
existing learning outcomes, assessment methods, as well as opportu­ 2.3. AI literacy in ECE
nities and challenges of AI literacy in ECE settings (Su & Yang, 2022).
AI education has posed challenges and opportunities to early child­
2. Literature review hood education (or kindergarten education), including why young
learners should learn AI in their early years, the subset of key AI con­
This scoping review provides a starting point for early childhood cepts that can be understood by children, and how children were
researchers and educators in terms of promoting research and practice engaged in a meaningful experience for them to acquire these concepts
related to AI literacy education. This section first reviews the current (Yang, 2022). He pointed out several reasons why young children need
status of both using AI, and the importance of AI literacy in early to learn AI: (1) knowing and understanding the basic competencies of AI
childhood education. and using AI applications is important for all citizens to become AI lit­
erates in today’s digital world (Ng et al., 2021a, b); (2) children need to
be empowered to understand, use, and evaluate AI with purposeful

2
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

guidance (Williams et al., 2019a); (3) children should have the capa­ systems, supervised machine learning, and generative AI through
bility to understand the basic functions of AI, especially when more Rock-Paper-Scissors game, Food Classification, Music Remix activities.
well-designed AI toys appear in their everyday experience (Kewalra­ Through these age-appropriate activities and tools, students could
mani et al., 2021). Su and Yang (2022) further identified AI in ECE recognise the basic knowledge about AI and understand the ethical is­
studies that have introduced AI concepts to kindergarteners using AI sues and limitations behind these tools. These studies provide the
learning tools such as PopBots and Zhorai. rationale and practices on how teachers can scaffold young children’s
Few prior researchers proposed what young children should learn understanding using these age-appropriate activities and tools (Ng et al.,
and how to scaffold them to understand how AI senses, perceives, in­ 2022c; Su & Yang, 2022; Yang, 2022).
teracts, behaves and creates. Yang (2022) designed an “AI for Kids”
Curriculum for young children to help them “highlight the role of 3. Methods
AI-powered technologies in human’s daily life and to enable children to
learn about AI using an embodied, project-based approach” through AI To our knowledge, there has been no existing review conducted to
toys, AI for Ocean at code.org and Quickdraw. Williams et al. (2019a) examine AI literacy in early childhood education. As such, the over­
designed an AI-interfaced robot for young children to help them arching goal of this review is to evaluate existing literature and identify
appreciate the abilities and limitations of AI agents and toys. Students challenges and opportunities of AI literacy in early childhood education.
can develop their relationships with these toys to learn knowledge-based The review approach is based on Arksey and O’Malley’s (2005)

Fig. 1. PRISMA diagram of included articles in the scoping review.

3
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

five-stage framework, which employs a rigorous transparency process to Table 1

improve the reliability of the research findings. The five steps of Arksey Methods Used in the Studies of Early AI literacy.
and O’Malley’s framework were used in this review of the literature on Studies Research Research Designs
early AI literacy: (1) identifying initial research questions, (2) identi­ Methods
fying the relevant studies, (3) selecting studies, (4) charting the data, Lin et al. (2020) Quantitative Assessments (Pre- and post- assessments)
and (5) collating. Kandlhofer et al. Qualitative Video data, pictures and observations
(2016) (field notes)
3.1. Identify the relevant studies Williams (2018) Quantitative Theory of mind assessment, rock paper
scissors assessment performance,
supervised learning assessment
According to Arksey and O’Malley (2005), a broad definition of performance, generative assessment
keywords for search terms should be used to obtain a ‘broad coverage’ of performance, pre- and post-tests of
existing literature. Keywords and search terms were investigated in children’s perception, attitudes
assessment
existing literature on early AI literacy from international perspectives.
Williams, Park, and Quantitative Knowledge assessments
The electronic databases used for the literature search included Web of Breazeal (2019)
Science, EBSCO, IEEE, ACM, Scopus, and Google Scholar. The search Druga et al. (2019) Quantitative Questionnaire
was limited to studies written in English. All articles are accessed in May Dwivedi et al. (2021) Mixed Machine learning metric and training
2022. To facilitate database search, this study investigated methods data
Druga and Ko (2021) Mixed Pre/post perception game responses and
peer-reviewed academic articles published until May 2022. We formu­ methods observations
lated a search string based on our understanding of and knowledge in Tseng et al. (2021) Quantitative Surveys
the AI in early childhood education domain and also, by referring to Kewalramani et al. Qualitative Interviews and observations
related AI education search strings used in other studies such as Su, Yim, (2021)
Druga et al. (2022) Qualitative Video recording: transcribed the videos
et al. (2022). The search string used for the review was: “AI literacy” OR
and noted comments
“Artificial intelligence literacy” AND “early childhood” OR “young Williams et al. Quantitative Assessments
child*” OR “preschool*” OR “kindergarten*” OR “pre-k*” OR “child­ (2019b)
care” OR “child care” OR “day care” OR “children”. Vartianinen et al. Qualitative Video recording
(2022)
Druga et al. (2022) Qualitative Interview
3.2. Study selection Tazume et al. (2020) Qualitative Observation

Using the author key search descriptors, 430 articles were identified.
As shown in Fig. 1, the Preferred Reporting Items for Systematic Reviews collection methods through collecting videos, pictures, field notes,
and Meta-Analyses (PRISMA) Statement was followed during the article comments, interviews, and observations. For example, Kandlhofer et al.
selection process (Moher et al., 2009). This paper focuses on artificial (2016) used observation and interview to assess children’s AI develop­
intelligence literacy in early childhood education. 430 articles were ment. Results show that they were very successful in introducing basic
found in the Web of Science, EBSCO, IEEE, ACM, Scopus, and Google AI/computer science concepts to children in a playful way. The rest of
Scholar [1 from Web of Science, 4 from EBSCO, 6 from IEEE, 1 from the papers used mixed research methods to collect data through various
ACM, 15 from Scopus and 403 from Google Scholar (n = 430)]. procedures, such as machine learning metric, training data, pre/post
The following articles were excluded from the study: the first perception game responses, and observations. Two articles (Su & Zhong,
exclusion criteria (E1) was articles that were not related to research title 2022; Yang, 2022), only introduced and described their AI curricula in
and abstract (n = 390). We excluded 390 results due to their irrelevance ECE, without implementing them in practice, and thus no data was
to the research topic: (a) whose focus was other topics, such as AI collected. More details are shown in Appendix 1.
technologies, medical, etc (n = 209); (b) whose focus was primary 3-6, Countries/Regions. This review shows that all studies were con­
secondary, and colleges/university students (n = 181). The second ducted in developed countries/regions (e.g., United States, Austria,
exclusion criteria (E2) was duplicate studies (n = 15). The third exclu­ Germany, Denmark, Sweden, Japan, Hong Kong, and Australia). As a
sion criteria (E3) was papers in which the participants or settings were result, we conclude that this type of article is strong enough to provide
not 3–8 years old (n = 1). Furthermore, the fourth exclusion criteria (E4) effective AI literacy research in ECE fields, representing AI educational
was papers whose focus/topic was not early artificial intelligence liter­ articles from various countries. More characteristics of the included
acy (n = 7). The last exclusion criteria (E5) was papers that were non- studies, such as research design, curriculum design, participants, tools,
English language studies (n = 1). and country can be found in Appendix 1.

3.3. Data charting and collation 4. Classification and taxonomy of AI literacy in ECE settings

The charting of selected articles is the last step in Arksey and According to Ng et al. (2021a, b), AI literacy instructional design can
O’Malley’s (2005) review framework. The 16 articles that focused on AI be developed in terms of pedagogy, content knowledge, tools and
literacy in ECE were thoroughly reviewed (2016, 1 article; 2018, 1 assessment methods used. This review examined instructional design,
article; 2019, 3 articles; 2020, 2 articles; 2021, 4 articles, and 2022, 5 tools, assessment methods and added learning outcomes that we found
articles). in the past studies, as shown in Appendix 1, according to the three
Research design. The research designs adopted in AI literacy in research questions. This study first lists the instructional designs that
early childhood education studies are shown in Table 1. Most studies have been used in the past studies, including design-based research
used a quantitative design (6 articles) and qualitative design (6 articles), (DBR) approach and interaction with AI robots. Then, it summarised the
followed by a mixed-research design (2 articles). The six quantitative tools used in the AI literacy courses. The second RQ aims to examine the
studies applied various data collection procedures, such as knowledge assessment methods such as knowledge/skill-based assessment, ques­
and theory of mind skills assessments, questionnaires, and surveys. For tionnaire and observations. The third RQ aims to review how prior
example, Lin et al. (2020) used assessments (i.e., knowledge test) to studies examine young children’s knowledge, concepts, skills and atti­
examine children’s machine learning knowledge through Zhorai. Re­ tudes towards AI literacy.
sults show that children can understand machine learning concepts With the coding framework, the selected articles were analysed in
more easily using the tool. The four qualitative studies used various data terms of its definitions and technologies used. The text segments were

4
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 2
Coding framework.
Themes Sub-themes Explanation Samples

RQ1: Learning Learning content A list of AI knowledge and skills that was suggested to be appropriate for kindergarteners Kandlhofer et al. (2016); Williams
design (Kandlhofer et al., 2016). (2018); Lin et al. (2020)
Tools AI tools that enable students to visualize AI concepts and interact with physical artifacts Williams (2018); Williams et al.
through programmable bots with sensors and AI-driven functions. (2019a, b)
Pedagogical design Different pedagogies used in designing learning activities. The major three approaches Druga et al. (2019); Williams et al.
are activity-based learning, experiential learning and project-based learning. (2019a)
RQ2: Assessment Knowledge and skill To examine the knowledge and skill acquisition of students’ AI literacy (Lin et al., 2020). Williams (2018); Williams et al.
methods assessments (2019a,b)
Questionnaires To examine children’s perception of robots (Williams, 2018) Williams (2018); Williams et al.
(2019a,b); Druga et al. (2019)
Observations To document students’ interaction in the AI learning experience (Druga & Ko, 2021). Druga and Ko (2021)
RQ3: Learning Knowledge and skills The learning outcomes of children’s AI knowledge and skills. Lin et al. (2020); Williams (2018);
outcomes Knowledge: increased AI concepts/knowledge; Kewalramani et al. (2021)
Skills: increased inquiry skills (i.e., creative, emotional, and collaborative).

extracted and coded under a coding scheme (see Table 2) and re­ discusses existing works from the aspects of instructional design, tools,
searchers could discuss any disagreements resolved and reach a final assessment methods and learning outcomes. Meanwhile, this study
decision. After validating the coding processes and categorisation, the concentrates on the challenges and opportunities for AI to assist edu­
findings were then descriptively summarised according to its frequency cators or researchers in locating relevant and important information in
and identified themes. This paper focuses on the instructional design, ECE.
tools, assessment methods and learning outcomes which contributes Our review focused on key aspects of the challenges and opportu­
researchers, educators and policy makers to designing and improving nities of improving AI literacy in early childhood education. The
the existing AI education practices and curricula to best foster young following are three research questions to help guide our search.
children’s AI literacy. Therefore, this review study adopts the frame­
work of AI literacy in ECE studies in terms of learning design, assessment 5. Results
methods, and learning outcomes (Fig. 2).
The first and second authors are coded all included articles filled in a
RQ1. How do researchers help young children develop AI literacy in
table separately. Information about learning design, assessment
terms of instructional design and tools?
methods, and learning outcomes were filled in by researchers. Differ­
ences between the first two researchers’ responses were then compared This section summarises how researchers design their instruction to
and discussed, leading to the revision of tables. develop young children’s AI literacy in terms of learning content, tools,
length of intervention and pedagogical design.
4.1. The purpose of the study Learning content. First, it is important to review the learning con­
tents in the curriculum design in the studies, which offer a basis for
This study contributes to a timely review of the current development educators to design their curriculum and programs for their students.
of the AI literacy education and offers guidance to researchers who are Although kindergarteners may not understand the working principles
seeking suitable activities to engage kindergarteners (e.g., Su & Yang, behind the AI concepts, young children can attempt to understand these
2022; Yang, 2022). The review also provides a reference for policy­ concepts through playful experience such as knowledge-based systems,
makers, who must make critical decisions regarding future development supervised machine learning, and generative AI (Williams et al., 2019a).
to set the educational standards. The examination of prior research in What does AI know, teaching AI robots to conduct tasks, experiencing
the field also helps researchers to determine which subjects are of machine learning, and its ethical concerns (Lin et al., 2020). Based on
continuing importance. There are few studies in the fields of AI literacy the suggested learning contents, future educators can know what
education in ECE settings. In these studies, researchers focus on topics knowledge content is suitable for young children to learn and experience
such as pedagogies, content knowledge, and technologies to uncover as shown in Table 3.
what, why and how to teach AI literacy for young learners (Yang, 2022). AI tools. In our review, 14 out of 16 studies engage young children to
But reviews of research on AI literacy in ECE are less common and this use AI tools/platforms to learn AI, as shown in Table 4. Among the
seems to be no review studies summarizing how educators develop studies, three of them used PopBots to foster kindergarten children’s
young children literacy in terms of age-appropriate instructional design basic AI understandings and concepts such as knowledge-based systems,
and tools, and what types of assessment methods have been used to supervised machine learning, and generative AI (Williams, 2018; Wil­
examine their AI literacy in the early AI curriculum, as well as its liams et al., 2019a, b). Other AI tools/platforms reported in the studies
learning outcomes throughout the interventions in ECE settings. To fill include Jibo robot, Anki’s Cozmo robot, Amazon’s Alexa, Zhorai, Cog­
these gaps in the literature, we present a scoping review of the literature nimates AI platform, Google’s Teachable Machine, Cosmo, Blue Bot, Coji
on AI literacy in ECE. The present study systematically analyses and by Wowee, Qobo the snail, and vernie-Lego Boost Bot. Furthermore, one

Fig. 2. AI Literacy in early childhood education.

5
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

study found that a conversational platform (Zhorai) can help children to Regarding the assessment methods used for evaluating young chil­
understand machine learning concepts (Lin et al., 2020). Teaching dren’s AI knowledge and skills, three data collection techniques,
young children AI was not possible in the past. However, with these including knowledge and theory of mind assessments (4 articles),
tools, kindergarten teachers start to explore ways to engage students to questionnaires and observation (3 articles), as shown in Table 5. It
experience AI which aims to enhance their technological skills that can should be noted that although there were some overlaps between
facilitate their learning and living. More characteristics of the AI tools, different assessment methods, this paper quotes the names used in the
such as categories, information, details, and country, can be found in reviewed papers directly.
Table 5. Different assessment methods were employed for examining young
Length of learning programs. Overall, the learning journey is quite children’s learning effects, such as knowledge and theory of mind skills
short for students to learn AI as shown in Table 3. Only two of the studies assessment. Particular knowledge assessments were designed and
have a longer intervention over weeks (Druga & Ko, 2021; Druga et al., developed for evaluating young children’s AI knowledge (Table 6). For
2022). Six studies engage students in short learning programs within example, a number of studies used knowledge assessments to assess
1–4 h (e.g., Williams et al., 2019a, b; Lin et al., 2020). Future studies are children’s AI concepts/knowledge (Lin et al., 2020; Williams, 2018;
necessary to consider longer interventions to foster students’ AI literacy. Williams et al., 2019a, b) (see Table 4). For example, Williams et al.
Pedagogy design. In terms of the pedagogy design used, studies (2019a,b) and Williams (2018) used three activities (i.e., Rock-­
have used various approaches. The two most commonly used pedagogies Paper-Scissors, Food Classification, Music Remix activities) to evaluate
are activity-based learning (n = 4) and experiential learning (n = 3). children’s different AI knowledge (i.e., knowledge-based systems, su­
“The activity-based approach has constructivist aims, and exploits both pervised ML, and generative AI). Results show that AI curriculum helped
collaborative interaction and access to information-rich resources” children improve three concepts, such as knowledge-based systems,
(Macdonald & Twining, 2002, p. 604). Firstly, researchers found that supervised ML, and generative AI.
hands-on activities could help students explore AI concepts effectively The questionnaires included AI perception questionnaire and
(e.g., Druga et al., 2019; Williams, 2018, Williams et al., 2019a). perception of robots questionnaire (Table 7). The AI perception ques­
Through hands-on experience, students can experience how AI works in tionnaires include several dimensions, including intelligence attribu­
robots and toys (i.e., knowledge-based systems, supervised machine tion, truthfulness attribution, and perceived understanding (Druga et al.,
learning, generative AI) in order to enhance children’s AI literacy 2019). Druga et al. (2019) developed an AI perception questionnaire for
(Williams, 2018; Williams et al., 2019a). Secondly, experiential learning assessing children’s feelings about the agents (i.e., Jibo robot, Anki’s
encourages students to learn by doing through hands-on experiences and Cozmo robot and Amazon’s Alexa, home assistant). The researchers
reflection that connect theories and knowledge learned in the classroom found that 68% of children thought the agents understood them the
to real-world situations (Morris, 2020). In our review, Kewalramani most.
et al. (2021) designed experiential AI learning activities through Different observations were employed for examining children’s
engaging 4-5-year-old children in playing with Vernie robots, thereby learning effects, such as observations of learning activities, young chil­
increasing children’s three literacy skills (i.e., create inquiry literacy, dren’s and educators’ play experiences, and interaction with AI robotics
emotional inquiry literacy, and collaborative inquiry literacy). Children (Table 8). For example, some researchers designed one activity about
playing with AI toys can explore their creative thinking and under­ interacting with an AI robot (RoBoHoN) and used observation to observe
standing of robots (creative inquiry literacy). Creative inquiry literacy children’s interaction with RoBonHoN. Results show that children
means “throughout their play with AI toys, the children were exposed to engaged in experiences that promoted the development of non-cognitive
many mediums through which they could explore and express their abilities” (Tazume et al., 2020, p. 323). “Interactions with AI robots are
creative thinking and understand-ing of robots” (Kewalramani et al., more similar to human communication than to communication with
2021, p.659). The two pedagogies could enable students to learn and other media, and this may affect the development of non-cognitive
experience AI although it is not necessary for young learners to know abilities” (Tazume et al., 2020, p. 328).
and understand the working principles behind AI.
RQ3. What were the learning outcomes of AI literacy in ECE settings?
Lastly, only a few researchers (n = 2) in the AI field have designed
projects to improve young children’s higher-order thinking skills (e.g., Most studies revealed that the AI curriculum was effective in terms of
communication, collaboration, and computational thinking skills) enhancing young children’s knowledge and skills, which facilitate
(Druga et al., 2019; Tseng et al., 2021). For example, Druga et al. (2019) young children to use AI tools for learning and living purposes. This
designed play and talk to AI agents (i.e., Alexa home assistant, Jibo and section summarised the learning outcomes in terms of knowledge gain
Cozmo robots) activities and children were invited to draw AI agents in and skill acquisition (see Table 9).
the future. It is understandable that young children may not be able to In terms of knowledge gain, AI curricula have significantly improved
learn much AI concepts and knowledge, but it is important to facilitate children’s AI concepts, such as knowledge-based systems, supervised
young students to use AI to communicate and collaborate with other ML, and generative AI (Williams, 2018; Williams et al., 2019a,b) and
classmates. To sum up, educators can consider the three pedagogical knowledge about machine learning (Lin et al., 2020; Tseng et al., 2021).
approaches (i.e., activity-based learning, experiential learning, and It was discovered that children can understand three AI concepts, such
hands-on experience) to encourage young children to experience how AI as rule-based systems, supervised machine learning, and generative AI
works, know how to use AI tools, communicate and collaborate with after basic AI courses. After online workshops, children know the ma­
others using AI, which facilitate their learning and living in today’s chine learning concepts/knowledge (Tseng et al., 2021). Moreover,
digital world. Activity-based learning is based on “the constructivist young children can improve their skills, such as theory of mind skills
learning approach, supports children throughout the learning process (Williams, 2018; Williams et al., 2019a,b) and creative inquiry,
with activities they do themselves” (Metin, 2020, p. 151). Experiential emotional inquiry and collaborative inquiry (Kewalramani et al., 2021).
learning means “knowledge creation resulting from reflection on and For example, Kewalramani et al. (2021) designed interactive learning
during experience, and its acknowledgement of the importance of activities among children to play with robots. Results show that children
learning processes and not simply behavioural outcomes” (Kolb, 2014, increased creative inquiry, emotional inquiry and collaborative inquiry
p. 26). Hands-on learning refers to learning by doing (Niiranen, 2021). after the AI activities. However, none of the research discusses how AI
Hands-on activities are important in children’s learning (Kim, 2020). learning programs affect young children’s affective learning outcomes
such as attitudes, motivation, and confidence of using AI. Future
RQ2. What were the assessment methods to examine young children’s
research is necessary to examine how students gain their affective
AI literacy in the early AI curriculum?
learning outcomes.

6
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 3
AI literacy in ECE studies.
Author/Year Title Learning content Tools Length of intervention Pedagogy design

Kandlhofer Artificial Intelligence and Computer Ten u nits: AI/computer / / /

et al. (2016) Science in Education: From science topic (e.g., principles
Kindergarten to University and concepts of AI, computer
science and robotics)
Williams PopBots Leveraging Social Robots to Rule-based systems, PopBots 60 min Activity-based learning:
(2018) Aid Preschool Children’s Artificial Supervised machine Rule-based systems;
Intelligence Education learning, and generative AI Supervised machine learning;
Generative AI
Druga et al. Inclusive AI literacy for kids around AI perception and Jibo robot, Anki’s 120 min Experiential learning:
(2019) the world expectations Cozmo robot and Interact with AI robots
Amazon’s Alexa
Williams et al. A is for Artificial Intelligence The Knowledge- based systems, PopBots 60 min Activity-based learning:
(2019a) Impact of Artificial Intelligence supervised machine learning, Knowledge-based systems;
Activities on Young Children’s and generative AI Supervised machine learning;
Perceptions of Robots Generative AI
Williams et al. PopBots: Designing an Artificial Knowledge- based systems, PopBots 45 min Activity-based learning:
(2019b) Intelligence Curriculum for Early supervised machine learning, Knowledge-based systems;
Childhood Education and generative AI Supervised machine learning;
Generative AI
Lin et al. (2020) Zhorai: Designing a Conversational Module 1: What does Zhorai Zhorai 240 min Experiential learning: Spoke
Agent for Children to Explore Know; Module 2: Teaching to Zhorai.
Machine Learning Concepts Zhorai;
Module 3: Witnessing
Machine Learning;
Module 4: AI and Ethics
Druga and Ko How do children’s perceptions of Train, code and test a series Cognimates AI 4-week Activity-based learning:
(2021) machine intelligence change when of smart programs platform “Make me Happy Program”,
training and coding smart “Rock Paper Scissors
programs? Program”, and “Smart Home
Program”.
Dwivedi et al. Introducing Children to Machine Machine learning concepts Teachable Machine / /
(2021) Learning Through Machine
Teaching
Tseng et al. PlushPal: Storytelling with The fundamentals of PlushPal 270 min Project-based learning:
(2021) Interactive Plush Toys and Machine machine learning and data Created a project using three
Learning science common gestures: none,
jump, and running.
Kewalramani Using Artificial Intelligence (AI)- Creative inquiry, emotional Cosmo, Blue Bot, Coji / Children play with Vernie
et al. (2021) interfaced robotic toys in early inquiry and collaborative by Wowee, Qobo the robots.
childhood settings: a case for inquiry snail, and vernie- Lego
children’s inquiry literacy Boost Bot
Druga et al. Family as a Third Space for AI Module 1: Image / 35 h Project-based learning
(2022) Literacies: How do children and classification;
parents learn about AI together? Module 2: Object
recognition;
Module 3: Voice assistants;
Module 4: Unplugged AI
games and co-design;
Module 5: Reflection on
study and learning activities
Su and Zhong Artificial intelligence (AI) in early Module 1: PictoBlox 370 min /
(2022) childhood education: Curriculum Introduction to AI;
design and future directions Module 2:
Machine learning;
Module 3:
Speech recognition;
Module 4
Flaws and biases of AI
Vartianinen Learning machine learning with 1.Interaction of children with Teachable machine / /
et al. (2022) very young children: who is teaching a Teachable Machine;
whom? 2. Explore the input–output
relationships with GTM
Yang (2022) Artificial Intelligence education for AI + Ocean Protection Popbo; Clearbot / /
young children: Why, what, and how
in curriculum design and
implementation
Druga et al. Family as a Third Space for AI Session 1: Image Teachable Machine 5-week /
(2022) Literacies: How do children and classification;
parents learn about AI together? Session 2: Object recognition;
Session 3: Voice assistants;
Session 4: Unplugged AI
games and co-design;
Session 5: Refection on study
and learning activities
(continued on next page)

7
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 3 (continued )
Author/Year Title Learning content Tools Length of intervention Pedagogy design

Tazume et al. Young children’s literacy and / RoBoHoN Results show that children Experiential learning:
(2020) cognition to interactive AI robots: A were strongly motivated to Interact with AI robots.
multifaceted study of potential participate in human-type
enhancement to early childhood AI media activities.
education

Table 4 Table 6
AI tools. Knowledge and Theory of Mind Assessment methods.
Learning Categories Information Details Country Studies Assessment Details Mode
tools methods

PopBots Hardware- https://www. Learn AI concepts USA Lin et al. (2020) Knowledge Knowledge assessment: e.g., Textual
focused media.mit.edu/ assessment What could you tell Zhorai
projects/ about monkeys so that it could
pop-kit/overvie correctly guess that monkeys
w/ live in rainforests?
Teachable Intelligent https://teacha Create machine USA Williams et al. Knowledge E.g., Tests if children Platform
Machine Agents blemachine. learning models (2019a,b); assessment understand how the robot
withgoogle.co Williams (PopBots) uses rules to decide who wins.
m/ (2018)
Zhorai Hardware- http://zhorai. Learn machine USA Williams (2018) Theory of mind Knowledge Assess: “Child sees Verbal
focused csail.mit.edu/ learning assessment what is in a box and judges
Jobo robot Hardware- https://jibo. Interact USA (yes - no) the knowledge of
focused com/ another person who does not
Anki’s Hardware- https://www. Interact USA see what is in a box.“;
Cozmo focused digitaldreamla Content False Belief: “Child
robot bs.com/pages judges another person’s false
/cozmo belief about what is in a
Amazon’s Hardware- https://dev Interact USA distinctive container when the
Alexa focused eloper.amazon. child knows what it is in the
com/en-US/a container.”
lexa Explicit False Belief: “Child
Cognimates Software or http:// Building games, USA judges how someone will
AI open source cognimates. programming search given that person’s
platform focused me/home/ robots, and mistaken belief.”
training AI models
PlushPal Software or https://www. Creating Japan
open source plushpal.app/ interactive toys
focused using the Table 7
Microbit! Questionnaire methods.
RoBoHoN Hardware- https://roboho Interact Japan
Studies Questionnaire Details Mode
focused n.com/global/
methods
PictoBlox Software or https://pict Programming India
open source oblox.ai/ Druga et al. AI perception If the agents are smarter Monster
focused (2019) questionnaire than they are; game
If the agents are truthful;
The agents understand
them the most
Williams et al. Perception of Do you agree with either Tablet
Table 5 (2019a); Robots view or are you in the computer or
Three data collection techniques commonly used. Williams Questionnaire middle? (1) Robots follow paper
(2018) rules/robots do not follow
Data Collection Techniques Studies
rules. (2) I am smarter
Assessments (i.e., Knowledge and theory of mind) Lin et al. (2020); than robots are smarter.
Williams (2018); (3) Robots can’t learn new
Williams et al. (2019a); things.
Willimas et al. (2019b) (4) Robots are like toys/
Questionnaires Druga et al. (2019); robots are like people.
Williams (2018); (5) Robots are like
Williams et al. (2019a) children/robots are like
Observations Druga and Ko (2021); adults.
Kewalramani et al. (2021);
Tazume et al. (2020)
early childhood education by addressing three research questions.
Based on the examination of AI curriculum designs in the reviewed
6. Discussion studies, several findings were identified. Firstly, most studies used age-
appropriate learning tools or platforms to enhance young children’s
This paper provides a review of AI literacy studies in early childhood learning AI. Robotic kits, especially PopBots, is the most common tool to
education, focusing on the characteristics and features of AI literacy as enhance children’s basic AI concepts (n = 3) (Williams, 2018; Williams
well as the effectiveness of interventions. Although there were few et al., 2019a,b).
empirical studies on AI literacy for ECE, the existing references did Secondly, in terms of pedagogy design, researchers design different
provide new insights into various aspects of AI literacy for young chil­ learning activities to increase children’s AI knowledge and concepts
dren. Our work contributes several new insights into AI literacies in (Williams, 2018; Williams et al., 2019a,b; Kandlhofer et al., 2016; Lin

8
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 8 Table 9
Observations method. Outcomes in AI literacy in ECE studies.
Studies Observations Activities Analyze Methods Author/Year Title Outcomes Theme
methods
Kandlhofer Artificial Intelligence Kindergarten children Knowledge
Druga and Ko Observations of “Make me Happy List of codes used for et al. (2016) and Computer Science investigated basic AI
(2021) learning Program”; transcripts analysis in Education: From topics and grasped
activities “Rock Paper from the different study Kindergarten to (simplified) AI
Scissors sites: University concepts in an
Program”; The first column: interesting way.
“Smart Home Initial Hypotheses, Test Williams PopBots Leveraging Kindergarten children Knowledge
Program" Assumptions, and (2018) Social Robots to Aid demonstrated
Refined Hypotheses; Preschool Children’s understanding of rule-
The second column: Artificial Intelligence based systems as well
Code: (1)Social Education as supervised
Judgement, Funds of machine learning
Knowledge, Egocentric, using a k-nearest
Observational, Agency neighbours algorithm.
(2) Edge Cases, Children do not
Common Cases, and understand
Agency; generative Al.
(3) Post-test Behavior, Druga et al. Inclusive AI literacy Young children in Knowledge
Social Intelligence, (2019) for kids around the high-income schools
Programmability, AI world and centres
Training. demonstrated a better
The third column: understanding of AI
Definition; concepts. Social,
The fourth column: economic, and
Study location cultural factors
Kewalramani Children’s and Play with the Video observation: play influenced AI
et al. (2021) educators’ play robotic toys with robotics toys, perception and
experiences (COSMO, Blue informal conversations expectations.
Bot, Coji by and interviews. Williams et al. A is for Artificial Kindergarten children Knowledge
Wowee, Qobo the Theme: Creative inquiry (2019a) Intelligence had the best
snail, and Vernie literacy, emotional The Impact of understanding of
- LegoBoost Bot) inquiry literacy, and Artificial Intelligence knowledge-based
collaborative inquiry Activities on Young systems.
literacy Children’s Perceptions
Tazume et al. Interact with Human-robot 1.Analyze using a text- of Robots
(2020) the AI robotics interaction mining technique: Williams et al. PopBots: Designing an Kindergarten children Knowledge
ranking of frequently (2019b) Artificial Intelligence are able to understand
used words and Co- Curriculum for Early AI concepts.
occurrence network Childhood Education
analysis of changes in Lin et al. Zhorai: Designing a Young children can Knowledge
utterance content; (2020) Conversational Agent understand machine
2. Categories and for Children to Explore learning concepts by
number of responses of Machine Learning interacting with
how children want to Concepts Zhorai and
play with RoBoHoN; completing the AI
3. Content of each modules.
kindergarten activity Druga and Ko How do children’s While training, Skills
(2021) perceptions of coding, and testing
machine intelligence their smart programs,
et al., 2020; Druga & Ko, 2021). For example, Williams (2018) and change when training children use the
and coding smart scientific method.
Williams et al. (2019a) designed three learning activities to promote
programs
children’s three basic AI concepts, such as knowledge-based systems, Dwivedi et al. Introducing Children Teaching machine Knowledge
supervised machine learning, and generative AI. Most studies found that (2021) to Machine Learning learning to children
AI education in kindergarten, primary 3-6, secondary, and higher edu­ Through Machine using interactive
cation are very different. Children ages 3–8 years old only understand Teaching machine learning
interfaces (teachable
basic AI concepts (Su & Zhong, 2022).
machines).
Thirdly, for young children’s learning outcomes, most of the studies Tseng et al. PlushPal: Storytelling PlushPal was Knowledge
were focused on examining young children’s improvement in terms of AI (2021) with Interactive Plush supposed to teach
or machine learning knowledge after their participation in the inter­ Toys and Machine children the
vention. In addition to AI or machine learning knowledge, children’s Learning fundamentals of
machine learning and
skills including creative inquiry, emotional inquiry, and collaborative data science.
inquiry, were improved after learning AI curricula (Kewalramani et al., Kewalramani Using Artificial Develop kindergarten Skills
2021; Williams et al., 2019a). Findings show that children (3–8 years et al. (2021) Intelligence (AI)- children’s inquiry
old) only understand basic AI concepts (e.g., Williams, 2018; Lin et al., interfaced robotic toys literacy using
in early childhood interactive AI-
2020). For example, Williams (2018) found that children understand
settings: a case for interfaced robotic
three basic AI concepts: knowledge-based systems, supervised machine children’s inquiry toys.
learning, generative AI. Furthermore, Kewalramani et al. (2021) applied literacy
a design-based research (DBR) approach to refine the instructional de­ Druga et al. Family as a Third Families with diverse Knowledge;
signs through planning, scaffolding, building children’s inquiry skills, (2022) Space for AI Literacies: AI perceptions, Skills
How do children and attitudes, and
teacher-child and peer-peer interactions, and assessment/review of
(continued on next page)
children’s learning for developing children’s inquiry. This approach is

9
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 9 (continued ) childhood education, more young children use AI technologies such as
Author/Year Title Outcomes Theme evaluation systems, educational robots, chatbots were incorporated to
enable teachers and parents to facilitate students’ cognitive and social
parents learn about AI knowledge can
together? successfully develop
development (Jin, 2019; Lin et al., 2020). These tools can help
AI literacy. increasingly be used to automate grading, interact with children and
Su and Zhong Artificial intelligence This paper suggests Knowledge; feedback activities. We can see that AI has become virtual teachers,
(2022) (AI) in early childhood problem-based Skills learning companions and caretakers to provide cognitive and social
education: Curriculum learning for AI
support for young children. However, young children who are the end
design and future education.
directions users of these tools may not know the technologies behind. They may
Vartiainen Learning machine Young children have Knowledge; have misconceptions about AI concepts and do not know the limitations
et al. (2022) learning with very successfully learned Skills and impacts behind.
young children: who is Google’s teachable Therefore, this review contributes to examine how to develop young
teaching whom? machine.
Yang (2022) Artificial Intelligence “Why”, “what”, and Knowledge
children’s AI literacy through instructional design and tools, and
education for young “how” for AI assessment methods. We also investigate how these learning programs
children: Why, what, education in early enhance young children’s learning outcomes in ECE settings. On top of
and how in curriculum childhood education these elements, the next section outlines the challenges and opportu­
design and
nities of AI literacy in early childhood education (Table 10).
implementation
Druga et al. Family as a Third Young children and Knowledge; Lack of teachers’ AI knowledge and skills. Teacher AI knowledge, in
(2022) Space for AI Literacies: parents learning Skills particular, was discovered to be one of the challenges encountered by
How do children and about AI technologies educators who do not have technical background, but need to develop
parents learn about AI were successful. and implement early AI literacy curriculum in their schools. It is a
together?
Tazume et al. Young children’s Young children’s Skills
challenge for most teachers. However, universities started to offer a
(2020) literacy and cognition media literacy and the wide range of AI activities for students such as the AI-for-kids curriculum
to interactive AI development of in Hong Kong, and developed professional training for ECE teachers.
robots: A multifaceted social-emotional skills They can make use of the resources to develop an age-appropriate cur­
study of potential (non-cognitive skills)
riculum and activities for their students (Laato et al., 2020). Future
enhancement to early may be impacted by
childhood education interactions with AI studies should explore what types of learning activities and tools are
robots. more suitable for early childhood education, and what kind of teacher
training is necessary for kindergarteners.
Low teachers’ confidence. Since most ECE teachers do not have a
appropriate for “small-scale educational research projects involving technical background and have not yet received related training, they
collaboration among educators, students, and researchers” (p. 56) to tend to have a low confidence level to understand AI, and a negative
enable teachers to improve their instructional design to best foster stu­ attitude to develop related activities and curriculum for their students.
dents’ AI literacy (Jetnikoff, 2015). Educators reported feeling overwhelmed by the entire process of
Fourthly, three assessment methods were adopted to assess chil­ experimenting with new digital tools (Dickey, 2011; Horton & Horton,
dren’s AI literacy in ECE, including knowledge and theory of mind skills 2003) and having little patience for dealing with minor technical issues
assessments (4 articles), questionnaires (3 articles), and observation (3 (Davidson, 2012; Drexler et al., 2008; Perry, 2015). This is worsened by
articles). We suggest further studies adopt performance-based mea­ teachers’ tight teaching schedule and they do not have any extra time to
surements for evaluating children’s AI knowledge/skills. Moreover, so learn AI knowledge. They may worry excessively since most of the
far there has been no standard questionnaire, survey or test for assessing currently available AI tools are children’s toys designed for children to
young children’s AI knowledge/skills. explore what AI is (Dyrbye et al., 2009). As such, teachers’ unreadiness
Finally, the current research on AI literacy in ECE classrooms lacks towards AI teaching is a great barrier for developing appropriate AI
empirical research from implementation and research methods not very materials and activities. However, some university and company may
rigorous. More empirical and interventional research designs are ex­ help develop some AI materials (Ng et al., 2023).
pected to be implemented by researchers in future, with clearly defined Lack of curriculum design. Few articles explain what, why, how, and
curriculum and control groups and a variety of data analysis techniques when children should learn. Less research on early AI education, though,
(e.g., t-tests, ANOVA). has concentrated on curriculum design (Su & Yang, 2022). Therefore,
Although included studies have discussed the positive cognitive curriculum design is the second of the challenges encountered by edu­
learning gain that these learning programs could offer students, some cators when engaging in the development and implementation of early
studies stated several obstacles when students learn AI. For instance, low AI literacy.
socioeconomic status (SES) children perform on average worse on AI Lack of teaching guidelines. Experiential exposure and project-based
abilities than children with higher SES backgrounds (Druga et al., 2019). learning is appropriate for children’s development of AI concepts
This situation also happens in AI literacy education. A study investi­ (Kandlhofer et al., 2016; Williams, 2018; Williams et al., 2019). Among
gating whether SES will impact children’s AI perception shows that AI these activities, most of the studies have identified that children scaffold
perception could be influenced by parents’ social-economical back­ their understandings through age-appropriate AI tools (e.g., Williams,
ground. The authors also found that “children in low and medium SES 2018; Williams et al., 2019); however, not all kindergartens have the
schools and centres tend to have stronger collaboration skills compared
to high SES children, but had a harder time advancing because they had
less experience with coding and interacting with these technologies” Table 10
(Druga et al., 2019, p. 1). Challenges and opportunities of early AI literacy.
Challenges N Sample studies
6.1. Challenges and opportunities of AI literacy in early childhood Lack of teachers’ AI knowledge, skills, confidence, and 10 Su and Zhong
education attitudes (2022)
Lack of curriculum design 6 Yang (2022)
Challenges Lack of teaching guidelines 5 Su and Zhong
(2022)
Current situation of using AI in early childhood education. In early

10
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Table 11
A framework of the elaborated elements of AI literacy for young children (Ng et al., 2021).
Elements Descriptions Examples

AI concepts Technical and conceptual understanding of the basic functions of AI Learning about classification, patterns and rules
AI practices The techniques and strategies used when applying AI Using tools to facilitate their everyday living and studies
AI perspectives Attitudes and dispositions adopted while solving problems Collaboration, communication, problem-solving, curiosity, inquiry

resources for children. Therefore, governments’ support is important to ● Researchers should design adaptable AI curriculum approaches, such
sponsoring schools to equip them with these technologies to enrich as holistic approaches. The six important components (i.e., AI
students’ exposures. Further, some websites (e.g., Machine learning for knowledge, AI processes, the impact of AI (content and product),
kids https://machinelearningforkids.co.uk/ and IBM https://www.ibm. student relevance, teacher-student communication, and flexibility
org/activities/machine-learning-for-kids) offer open-source pre­ (process and praxis) for designing AI curriculum using holistic
school-oriented AI curriculum using free web-based software such as method (Chiu, 2021).
Teachable Machine for young children to learn AI (Machine learning for ● Educators should design more interesting AI activities for young
kids, n.d.; IBM.org, n.d.). children, which could change their existing attitudes towards AI.
● A technological, pedagogical, and content knowledge (TPACK)
Opportunities framework should be taken into consideration to provide a roadmap
Developing AI concepts and practices. In the past, it was challenging for understanding how to effectively integrate AI literacy into
to foster young children to learn AI concepts and skills due to a lack of classrooms in order to aid educators in teaching more effectively (Ng
suitable tools. With a growing number of developmentally well-designed et al., 2022b).
technologies for young children, young children can explore what AI is ● Future researchers and educators will develop quantitative and
through playful experiences. For example, young children can play with qualitative assessments to assess children’s learning outcome
AI social robots, toys and services (e.g., Cozmo, PopBots, Quickdraw, through pre and post-knowledge tests, surveys, and observations.
Wowee) to learn how to interact with AI tools that make responses based
on children’s emotions and expressions and stimulate their curiosity to 6.3. Pedagogical implications
explore AI technologies in their everyday lives (Kewalramani et al.,
2021; McStay & Rosner, 2021). The interaction with AI can help stu­ In order to help young children develop AI literacy skills at an ECE
dents improve their digital literacy skills and attitudes, preparing them level, this paper first introduces various teaching content, tools, and
better for their progression in primary schools even though it is pedagogy that are appropriate for this age group. This can aid in the
acknowledged that young children at this age hardly know and under­ development of age-appropriate AI curriculum, games, and toys for kids
stand AI. Parents could use these AI toys and services to create a digital by researchers and educators (Ng & Chu, 2021). This article synthesises
environment for their kids, and this new skill may make them more the evidence that shows the effectiveness of the AI curriculum in ECE.
comfortable with the tool, increasing the likelihood that they will use it Most studies have shown that the AI curricula have significantly
in the future as part of their instructional design. For example, paren­ improved children’s AI or machine learning concepts (Williams et al.,
t–child interactions through AI toys (i.e., Alpha Mini, Coji, Qobo the 2019a,b; Ng et al., 2022a) and theory of mind skills and several inquiries
snail, LegoBoost Bot) activities to enhance children’s social and (i.e., creative inquiry, emotional inquiry and collaborative inquiry). One
emotional abilities (Kewalramani et al., 2021). study designed AI activities to enhance kindergarten children’s creative
Developing AI perspectives. AI perspectives refers to attitudes and inquiry, emotional inquiry and collaborative inquiry (Kewalramani
dispositions adopted while solving problems (Ng et al., 2021). Most et al., 2021). Moreover, teachers were able to effectively implement the
studies show that teaching young children early AI concepts and skills AI curricula in order to promote children’s learning and gains. For
significantly enhances their understanding of AI (e.g., Rule-based sys­ example, researchers could design interesting AI activities (e.g., interact
tems, Supervised machine learning, and generative AI) (Williams, 2018; with AI robots) for children to enhance children’s AI perceptions
Williams et al., 2019). However, children may not comprehend the (Tazume et al., 2020). Furthermore, by presenting the various assess­
knowledge and concepts behind AI at a young age because AI in ECE is ment methods used, the research findings in this paper advance AI ed­
vastly different from AI in primary and secondary schools. For example, ucation in ECE research. Three frequent assessment tools were most
AI in ECE mainly focuses on how to use AI tools (i.e., AI toys) to facilitate used, including knowledge and skill assessments, observation, and
their daily life and studies, but AI in primary and secondary schools questionnaires. For example, Williams (2018) used knowledge assess­
focuses on knowledge and skills acquisition. However, young children ments to examine children’s AI knowledge and skills. Results show that
can develop their digital skills through play and exploration, enhancing the AI curriculum for ECE has significantly improved children’s
their emotional, collaborative, and inquiry literacy as well as related AI perceived AI knowledge. Therefore, there is a need to promote quality AI
perspectives like social skills (like how to play and interact with other education in the early years at the fronts of pedagogical.
kids), curiosity, and others (Kewalramani et al., 2021). Children use AI This paper also outlines the challenges and opportunities of AI lit­
devices (e.g., drawing tools, robotic toys, chatbots) on a daily basis to eracy in early childhood education for researchers and educators as a
improve communication skills through group activities and stimulate reference. The challenges of AI literacy in ECE, including lack of
their imagination through digital storytelling and writing (Ng et al., teachers’ AI knowledge and skills, low teachers’ confidence, lack of
2022c). Based on the above, it is worth developing AI literacy for chil­ curriculum design, and lack of teaching guidelines. On the other hand,
dren in ECE. The following demonstrates a summary of AI concepts, we identified some opportunities for AI literacy in ECE, including
practices and perspectives adapted from Ng et al. (2021b), as shown in developing AI concepts and practices and AI perspectives. This paper
Table 11. provides some suggestions for researchers and educators. For example,
future educators could design some engaging AI activities for young
children’s learning about AI, such as interacting with AI robots, etc. Our
6.2. Suggestions
discussion provides a good starting point for scholars to further inves­
tigate the effectiveness of early AI literacy and for various stakeholders
To address teaching challenges, educators should update their AI
to develop stimulating and engaging AI courses in educational fields.
knowledge. Early childhood educators can use digital resources to learn
AI concepts in their spare time.

11
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

7. Conclusion research designs for AI in ECE settings. Secondly, The research’s scope
was constrained by the paucity of literature on AI literacy for early
This review contributes to mapping the research design, tools, childhood education, making it impossible to quantify generalisations.
research methods, intervention programs, and research findings However, according to the available literature, early AI literacy is a
ascribed to the existing studies on early AI curriculum. It also explains worthwhile investigation and a highly valuable emerging research di­
the challenges and opportunities of AI literacy in early childhood edu­ rection for future researchers.
cation. The results of this literature survey can inform future research in
terms of advancing tools, pedagogical methods, research design, Credit author statement
research methods, intervention, and assessment for early AI curricula
and provide researchers and practitioners a guide for the design, Jiahong Su: Conceptualization, Methodology, Formal analysis, Data
implementation, and evaluation of age-appropriate AI curricula for curation, Writing - Original draft preparation, Review & Editing, Visu­
young children. This review would provide valuable directions for early alization. Davy Ng Tsz Kit: Review & Editing. Samuel Chu Kai Wah:
AI education and serve as a reference for future ECE research in the Review & Editing.
digital society.
Two limitations in this study. Firstly, most of the studies selected in Declaration of competing interest
this review only are of their beginning stage and their research methods
were preliminary, descriptive and exploratory in nature. More re­ The authors declare that they have no known competing financial
searchers are anticipated to join the conversation, contributing more interests or personal relationships that could have appeared to influence
rigorous research techniques, as well as successful instructional and the work reported in this paper.

Appendix 1

AI curriculum for early childhood education.

Studies Research design Curriculum design Participants Tools Countries

Lin et al. (2020) Assessments (Pre- and post- assessments) Module 1: What does Zhorai 14 children Zhorai USA
Know; (Mean: 9.43
Module 2: Teaching Zhorai; years old)
Module 3: Witnessing
Machine Learning;
Module 4: AI and Ethics
Kandlhofer et al. Video data, pictures and observations (field notes) Ten units: AI/computer 24 children No mentioned Austria
(2016) science topic (Mean: 5 years
old)
Williams (2018) Theory of mind assessment, rock paper scissors Rule-based systems, 80 children (4–6 PopBots USA
assessment performance, supervised learning Supervised machine years old)
assessment performance, generative assessment learning, and generative AI
performance, pre- and post-tests of children’s
perception, attitudes assessment
Williams et al. Assessments knowledge- based systems, 80 children (4–6 PopBots USA
(2019a) supervised machine years old)
learning, and generative AI
Druga et al. Questionnaire AI perception and 102 children Jibo robot, Anki’s Cozmo U.S.A, Germany,
(2019) expectations (7–12 years old) robot and Amazon’s Alexa Denmark, and
Sweden
Dwivedi et al. Machine learning metric and training data Machine learning concepts 14 children Teachable Machine USA
(2021) (7–13 years old)
Druga and Ko Pre/post perception game responses and observations Cognimates AI platform 52 children Cognimates AI platform USA
(2021) (train, code and test a series (7–12 years old)
of smart programs)
Tseng et al. Surveys The fundamentals of 11 children PlushPal Japan
(2021) machine learning and data (8–14 years old)
science
Kewalramani Interviews and observations Creative inquiry, emotional 21 children (4–5 Cosmo, Blue Bot, Coji by Australia
et al. (2021) inquiry and collaborative years old) Wowee, Qobo the snail,
inquiry and vernie- Lego Boost Bot
Druga et al. Video recording: transcribed the videos and noted Module 1: Image 18 children / USA
(2022) comments classification; (5–11 years old)
Module 2: Object
recognition;
Module 3: Voice assistants;
Module 4: Unplugged AI
games and co-design;
Module 5: Reflection on
study and learning activities
Williams et al. Knowledge assessments Knowledge- based systems, 80 children PopBots USA
(2019b) supervised machine
learning, and generative AI
Yang (2022) / AI + Ocean Protection / Popbo; Clearbot Hong Kong
Video analysis 6 children Finland
(continued on next page)

12
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

(continued )
Studies Research design Curriculum design Participants Tools Countries

Vartianinen 1.Interaction of children Google’s teachable

et al. (2022) with a Teachable Machine; machine
2. Explore the input–output
relationships with GTM
Su and Zhong / Module: / PictoBlox Hong Kong
(2022) Introduction to AI;
Module 2:
Machine learning;
Module 3:
Speech recognition;
Module 4
Flaws and biases of AI
Druga et al. Interview Session 1: Image 18 children and / USA
(2022) classification; 16 parents
Session 2: Object
recognition;
Session 3: Voice assistants;
Session 4: Unplugged AI
games and co-design;
Session 5: Refection on
study and learning
activities.
Tazume et al. Observation Interactive with AI robot RoBoHoN Japan
(2020)

References Kandlhofer, M., Steinbauer, G., Hirschmugl-Gaisch, S., & Huber, P. (2016, October).
Artificial intelligence and computer science in education: From kindergarten to
university. In 2016 IEEE frontiers in education conference (FIE) (pp. 1–9). IEEE.
Arksey, H., & O’Malley, L. (2005). Scoping studies: Towards a methodological
Kewalramani, S., Palaiologou, I., Dardanou, M., Allen, K. A., & Phillipson, S. (2021).
framework. International Journal of Social Research Methodology, 8(1), 19–32.
Using robotic toys in early childhood education to support children’s social and
Burgsteiner, H., Kandlhofer, M., & Steinbauer, G. (2016, March). Irobot: Teaching the
emotional competencies. Australasian Journal of Early Childhood, 46(4), 355–369.
basics of artificial intelligence in high schools. Proceedings of the AAAI Conference on
Kim, J. (2020). Learning and teaching online during Covid-19: Experiences of student
Artificial Intelligence, 30(1), 4126–4127.
teachers in an early childhood education practicum. International Journal of Early
Chen, X., Xie, H., Zou, D., & Hwang, G. J. (2020). Application and theory gaps during the
Childhood, 52(2), 145–158.
rise of artificial intelligence in education. Computers & Education: Artificial
Kolb, D. A. (2014). Experiential learning: Experience as the source of learning and
Intelligence, 1, Article 100002.
development. FT press.
Chiu, T. K. (2021). A holistic approach to the design of artificial intelligence (AI)
Kong, S. C., Cheung, W. M. Y., & Zhang, G. (2021). Evaluation of an artificial intelligence
education for K-12 schools. TechTrends, 65(5), 796–807.
literacy course for university students with diverse study backgrounds. Computers &
Davidson, L. Y. J. (2012). Teachers’ perspective on using technology as an instructional tool.
Education: Artificial Intelligence, 2, Article 100026.
Doctoral dissertation, Walden University.
Kong, S. C., Cheung, W. M. Y., & Zhang, G. (2022). Evaluating artificial intelligence
Dickey, M. D. (2011). The pragmatics of virtual worlds for K-12 educators: Investigating
literacy courses for fostering conceptual learning, literacy and empowerment in
the affordances and constraints of Active Worlds and Second Life with K-12 in-
university students: Refocusing to conceptual building. Computers in Human Behavior
service teachers. Educational Technology Research & Development, 59(1), 1–20.
Reports, 7, Article 100223.
Drexler, W., Baralt, A., & Dawson, K. (2008). The teach web 2.0 consortium: A tool to
Kong, S. C., Cheung, W. M. Y., & Zhang, G. (2023). Evaluating an artificial intelligence
promote educational social networking and web 2.0 use among educators.
literacy programme for developing university students’ conceptual understanding,
Educational Media International, 45(4), 271–283.
literacy, empowerment and ethical awareness. Educational Technology & Society, 26
Druga, S., Christoph, F. L., & Ko, A. J. (2022, April). Family as a Third Space for AI
(1), 16–30.
Literacies: How do children and parents learn about AI together?. In CHI conference
Laato, S., Vilppu, H., Heimonen, J., Hakkala, A., Björne, J., Farooq, A., … Airola, A.
on human factors in computing systems (pp. 1–17).
(2020, October). Propagating AI knowledge across university disciplines-the design
Druga, S., & Ko, A. J. (2021, June). How do children’s perceptions of machine
of a multidisciplinary AI study module. In 2020 IEEE frontiers in education conference
intelligence change when training and coding smart programs?. In Interaction design
(FIE) (pp. 1–9). IEEE.
and children (pp. 49–61).
Liang, J. C., Hwang, G. J., Chen, M. R. A., & Darmawansah, D. (2021). Roles and research
Druga, S., Vu, S. T., Likhith, E., & Qiu, T. (2019). In Inclusive AI literacy for kids around the
foci of artificial intelligence in language education: An integrated bibliographic
world (pp. 104–111).
analysis and systematic review approach. Interactive Learning Environments, 1–27.
Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T., …
Lin, P., Van Brummelen, J., Lukin, G., Williams, R., & Breazeal, C. (2020, April). Zhorai:
Williams, M. D. (2021). Artificial Intelligence (AI): Multidisciplinary perspectives on
Designing a conversational agent for children to explore machine learning concepts.
emerging challenges, opportunities, and agenda for research, practice and policy.
Proceedings of the AAAI Conference on Artificial Intelligence, 34(9), 13381–13388.
International Journal of Information Management, 57, 101994.
Long, D., & Magerko, B. (2020, April). What is AI literacy? Competencies and design
Dyrbye, L., Cumyn, A., Day, H., & Heflin, M. (2009). A qualitative study of physicians’
considerations. In Proceedings of the 2020 CHI conference on human factors in
experiences with online learning in a masters degree program: Benefits, challenges,
computing systems (pp. 1–16).
and proposed solutions. Medical Teacher, 31(2), e40–e46.
Macdonald, J., & Twining, P. (2002). Assessing activity–based learning for a networked
Eguchi, A., Okada, H., & Muto, Y. (2021). Contextualizing AI education for K-12 students
course. British Journal of Educational Technology, 33(5), 603–618.
to enhance their learning of AI literacy through culturally responsive approaches. KI-
Machine learning for kids (n.d). Teach a computer to play a game https://machinelearni
Künstliche Intelligenz, 35(2), 153–161.
ngforkids.co.uk/.
Gaube, S., Suresh, H., Raue, M., Merritt, A., Berkowitz, S. J., Lermer, E., Couhlin, J.,
McCarthy, J. (2007). From here to human-level AI. Artificial Intelligence, 171(18),
Guttang, J., Colak, E., & Ghassemi, M. (2021). Do as AI say: Susceptibility in
1174–1182.
deployment of clinical decision-aids. NPJ digital medicine, 4(1), 1–8.
McStay, A., & Rosner, G. (2021). Emotional artificial intelligence in children’s toys and
Horton, W., & Horton, K. (2003). E-Learning tools and technologies: A consumer’s guide for
devices: Ethics, governance and practical remedies. Big Data & Society, 8(1), Article
trainers, teachers, educators, and instructional designers. John Wiley & Sons.
2053951721994877.
IBM. (n.d.). Machine Learning for kids. https://www.ibm.org/activities/machine-learn
Metin, S. (2020). Activity-based unplugged coding during the preschool period.
ing-for-kids.
International Journal of Technology and Design Education, 1–17.
Jetnikoff, A. (2015). Design based research methodology for teaching with technology in
Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & PRISMA Group*.. (2009). Preferred
English. English in Australia, 50(3), 56–60.
reporting items for systematic reviews and meta-analyses: The PRISMA statement.
Jin, L. (2019, August). Investigation on potential application of artificial intelligence in
Annals of Internal Medicine, 151(4), 264–269.
preschool children’s education. Journal of Physics: Conference Series. IOP Publishing,
Mondal, B. (2020). Artificial intelligence: State of the art. Recent Trends and Advances in
1288(1), 1–5.
Artificial Intelligence and Internet of Things, 389–425.

13
J. Su et al. Computers and Education: Artificial Intelligence 4 (2023) 100124

Morris, T. H. (2020). Experiential learning–a systematic review and revision of Kolb’s Su, J., & Zhong, Y. (2022). Artificial intelligence (AI) in early childhood education:
model. Interactive Learning Environments, 28(8), 1064–1077. Curriculum design and future directions. Computers & Education: Artificial
Mousavinasab, E., Zarifsanaiey, N., R., Niakan Kalhori, S., Rakhshan, M., Keikha, L., & Intelligence, Article 100072.
Ghazi Saeedi, M. (2021). Intelligent tutoring systems: A systematic review of Su, J., Zhong, Y., & Ng, D. T. K. (2022). A meta-review of literature on educational
characteristics, applications, and evaluation methods. Interactive Learning approaches for teaching AI at the K-12 levels in the Asia-Pacific region. Computers &
Environments, 29(1), 142–163. Education: Artificial Intelligence, Article 100065.
Nan, J. (2020, August). Research of application of artificial intelligence in preschool Tazume, H., Morita, T., & Hotta, H. (2020, June). Young children’s literacy and cognition
education. Journal of Physics: Conference Series. IOP Publishing, 1607(1), 1–5. to interactive AI robots: A multifaceted study of potential enhancement to early
Ng, D. T. K., & Chu, S. K. W. (2021). Motivating students to learn AI through social childhood education. In EdMedia+ innovate learning (pp. 323–328). Association for
networking sites: A case study in Hong Kong. Online Learning, 25(1), 195–208. the Advancement of Computing in Education (AACE).
Ng, D. T. K., Lee, M., Tan, R. J. Y., Hu, X., Downie, J. S., & Chu, S. K. W. (2022a). A review Touretzky, D., Gardner-McCune, C., Martin, F., & Seehorn, D. (2019, July). Envisioning
of AI teaching and learning from 2000 to 2020. Education and Information AI for K-12: What should every child know about AI? Proceedings of the AAAI
Technologies, 1-57. Conference on Artificial Intelligence, 33(1), 9795–9799.
Ng, D. T. K., Leung, J. K. L., Chu, K. W. S., & Qiao, M. S. (2021a). AI literacy: Definition, Tseng, T., Murai, Y., Freed, N., Gelosi, D., Ta, T. D., & Kawahara, Y. (2021, June).
teaching, evaluation and ethical issues. Proceedings of the Association for Information PlushPal: Storytelling with interactive plush toys and machine learning. In
Science and Technology, 58(1), 504–509. Interaction design and children (pp. 236–245).
Ng, D. T. K., Leung, J. K. L., Chu, S. K. W., & Qiao, M. S. (2021b). Conceptualizing AI Vartiainen, H., Tedre, M., & Valtonen, T. (2020). Learning machine learning with very
literacy: An exploratory review. Computers & Education: Artificial Intelligence, 2, young children: Who is teaching whom? International Journal of Child-Computer
Article 100041. Interaction, 25, 100182.
Ng, D. T. K., Leung, J. K. L., Su, J., Ng, R. C. W., & Chu, S. K. W. (2023). Teachers’ AI Williams, R. (2018). PopBots: Leveraging social robots to aid preschool children’s artificial
digital competencies and twenty-first century skills in the post-pandemic world. intelligence education. Doctoral dissertation, Massachusetts Institute of Technology.
Educational Technology Research and Development, 1-25. Williams, R., Park, H. W., & Breazeal, C. (2019, May). A is for artificial intelligence: the
Ng, D. T. K., Leung, J. K. L., Su, M. J., Yim, I. H. Y., Qiao, M. S., & Chu, S. K. W. (2022b). impact of artificial intelligence activities on young children’s perceptions of robots.
AI Literacy Education in Early Childhood Education. In AI Literacy in K-16 Classrooms In Proceedings of the 2019 CHI conference on human factors in computing systems, 1-11.
(pp. 63–74). Cham: Springer International Publishing. Williams, R., Park, H. W., Oh, L., & Breazeal, C. (2019, July). Popbots: Designing an
Ng, D. T. K., Luo, W. Y., Chan, H. M. Y., & Chu, S. K. W. (2022c). Using digital story artificial intelligence curriculum for early childhood education. Proceedings of the
writing as a pedagogy to develop AI literacy among primary students. Computers & AAAI Conference on Artificial Intelligence, 33(1), 9729–9736.
Education: Artificial Intelligence, Article 100054. Xu, Y., Liu, X., Cao, X., Huang, C., Liu, E., Qian, S., … Zhang, J. (2021). Artificial
Niiranen, S. (2021). Supporting the development of students’ technological intelligence: A powerful paradigm for scientific research. The Innovation, 2(4),
understanding in craft and technology education via the learning-by-doing Article 100179.
approach. International Journal of Technology and Design Education, 31(1), 81–93. Yang, W. (2022). Artificial Intelligence education for young children: Why, what, and
Perry, B. (2015). Gamifying French language learning: A case study examining a quest- how in curriculum design and implementation. Computers & Education: Artificial
based, augmented reality mobile learning-tool. Procedia-Social and Behavioral Intelligence, Article 100061.
Sciences, 174, 2308–2315. Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review
Steinbauer, G., Kandlhofer, M., Chklovski, T., Heintz, F., & Koenig, S. (2021). of research on artificial intelligence applications in higher education–where are the
A differentiated discussion about AI education K-12. KI-Künstliche Intelligenz, 35(2), educators? International Journal of Educational Technology in Higher Education, 16(1),
131–137. 1–27.
Su, J., & Yang, W. (2022). Artificial intelligence in early childhood education: A scoping Zheng, L., Niu, J., Zhong, L., & Gyasi, J. F. (2021). The effectiveness of artificial
review. Computers & Education: Artificial Intelligence, Article 100049. intelligence on learning achievement and learning perception: A meta-analysis.
Su, M. J., Yim, I. H. Y., Ng, D. T. K., Leung, K. L., & Chu, S. K. W. (2022b). AI literacy Interactive Learning Environments, 1–15.
education in kindergarten setting: A review. In International conference and institute
on AI and blockchain (ICAIB 2022) for information and library science, 1-13.

14