You are on page 1of 7

Computers and Education: Artificial Intelligence 2 (2021) 100006

Contents lists available at ScienceDirect

Computers and Education: Artificial Intelligence


journal homepage: www.elsevier.com/locate/caeai

Modeling the structural relationship among primary students’ motivation to


learn artificial intelligence
Pei-Yi Lin a, Ching-Sing Chai b, Morris Siu-Yung Jong b, *, Yun Dai b, Yanmei Guo c, Jianjun Qin d
a
Department of Education, National Kaohsiung Normal University, Taiwan
b
Department of Curriculum and Instruction, The Chinese University of Hong Kong, Hong Kong
c
Teacher Training and Development Centre of Dongcheng District Beijing, Beijing, China
d
School of Mechanical-Electronic and Vehicle Engineering, Beijing University of Civil Engineering and Architecture, Beijing, 100044, China

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

Keywords: The recent advances in artificial intelligence (AI) present both challenges and opportunities for educational
Artificial intelligence practitioners. A new AI curriculum has been developed and piloted in many primary schools in Beijing, China. The
ARCS present study had two aims: (1) to test the factor structure of students’ motivation to learn AI and (2) to examine
Intrinsic motivation
possible gender differences in students’ motivation to learn AI. This online questionnaire–based research recruited
Career motivation
Gender difference
420 primary students from the piloting schools. Structural equation modeling was employed to test a hypothe-
sized model comprising six motivational factors and strategies: (1) intrinsic motivation, (2) career motivation, (3)
attention, (4) relevance, (5) confidence, and (6) satisfaction. The study discovered intrinsic motivation to have the
strongest influence on career motivation, while the motivational strategies of attention, relevance, and confidence
also influenced career motivation. Additionally, compared with female students, male students scored higher in
terms of motivational factors and strategies. The findings serve as a reference for the future development of AI
curricula and instruction.

1. Introduction Currently, AI as a discipline and associated studies on learners’ per-


ceptions of learning AI is largely confined to computer science and sci-
The Oxford English Living Dictionary defines artificial intelligence ence, technology, engineering, and mathematics (STEM) departments in
(AI) as a branch of computer science that concerns computer systems that higher education institutes (Geng et al., 2019; Zawacki-Richter et al.,
can perform tasks that usually require human intelligence. As a booming 2019). However, given the growing importance of AI technology, edu-
technology, AI has been revolutionizing the economic and societal sec- cators are advocating for AI to be included in elementary and secondary
tors, including the field of education (Shang et al., 2006). AI de- school curricula (Knox, 2020). The associated curriculum questions that
velopments in education primarily include building adaptive systems, arise such as when to educate the young, how to design the curricula,
personalization, and intelligent tutoring systems (Zawacki-Richter et al., what to include for different levels of students, and the effects of such
2019). Recent AI research has introduced new teaching tools and decisions warrant further research. Additionally, AI-related elements
methods of learning that focus on providing human–AI interaction may cultivate teachers’ and students’ digital competency in an
(Starcic, 2019). Currently, ideas such as AI-assisted language learning (Fu AI-saturated world. Certain initiatives, such as promoting computational
et al., 2020), AI adaptive instructional guidance in a mixed-reality setting thinking, have been taken to expand K-12 students’ digital learning ex-
(Yannier et al., 2020), and pedagogical agents to foster students’ moti- periences, coding ability, and algorithmic problem-solving ability with
vation (Veletsianos and Russell, 2014) are being tested and launched. computers (So et al., 2020). However, AI learning in the K-12 setting is
The ongoing research on AI, however, has focused on systems develop- still in its nascent stage. Incorporating new technologies implies new
ment and application. There is a lack of studies adopting educational pedagogical ideas. It is suggested to evaluate levels of motivations among
framework to investigate students learning motivation among AI learners AI learners when incorporating AI-technologies in K-12 classroom setting
(Chen et al., 2020). (Hwang et al., 2020).

* Corresponding author.
E-mail addresses: pylin@nknu.edu.tw (P.-Y. Lin), cschai@cuhk.edu.hk (C.-S. Chai), mjong@cuhk.edu.hk (M.S.-Y. Jong), yundai@cuhk.edu.hk (Y. Dai), guoyanmei_
0907@163.com (Y. Guo), qinjianjun@bucea.edu.cn (J. Qin).

https://doi.org/10.1016/j.caeai.2020.100006
Received 25 October 2020; Received in revised form 12 December 2020; Accepted 13 December 2020
2666-920X/© 2020 The Author(s). 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/).
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

Motivation is directly related to the development of students’ atti- they are curious about it or find the topic interesting. AI is a novel
tudes and persistent efforts toward achieving a goal (Ryan and Deci, advancement, and news such as AI beating human experts in the Go game
2000). It is an important determinant of learning performances and is intriguing to many. However, developing students’ initial interest in AI
career decisions. Therefore, this study explored students’ intrinsic through sound instructional design is critical to increase students’
motivation that drives their participation in AI learning for reasons such knowledge of the subject matter and to help them to decide whether they
as challenge, curiosity, and mastery. Ideally, harnessing students’ should continue learning the subject in the long run. Students must invest
intrinsic motivation during instruction can facilitate their career moti- sustained efforts to confidently master the increasingly complex knowl-
vation in the long term, because people tend to choose a career based on edge in the field of AI. Additionally, they must continuously assess the
their inherent interests and the value they attach to a career. However, subject’s relevance to their career goals (Authors, in press). When stu-
studies have reported that students’ intrinsic motivation to engage in dents’ initial interest is enhanced through a well-designed curriculum,
school-related activities and set career goals declines when they advance their initial motivation may likely evolve into a long-term career choice.
to higher grades (Scherrer and Preckel, 2019). Therefore, to sustain
students’ learning interest in the long term, curriculum designers must 2.2. Motivational strategies to learn AI: the ARCS model approach
address students’ motivation needs (e.g., strategies to capture students’
attention and engage them in learning tasks) (Ryan and Deci, 2020). The In terms of pedagogy, the ARCS model provides instructors with
successful application of Keller’s (1987) motivational concepts of atten- guidelines for designing and developing strategies to motivate student
tion, relevance, confidence, and satisfaction (ARCS) across numerous learning (Keller and Suzuki, 2004; Wah, 2015). The model requires
instructional settings highlight the positive influence student motivation curriculum designers to adopt appropriate instructional strategies to
has on their learning (for example, see Ucar & Kumtepe [2020] and Jong attract and maintain students’ attention to the instruction because
[2020]; for a review, see Li & Keller [2018]). Hence, the ARCS motiva- attention is a prerequisite for information processing, which in turn leads
tional model has the potential to serve as a pedagogical framework for to understanding. Understanding a subject could build students’ confi-
delivering satisfying learning experiences, thereby sustaining students’ dence and help them recognize the relevance of the subject matter to
motivation to learn AI. Additionally, this sustained motivation may lead their goals. These three factors (i.e., attention, relevance, and confidence)
to some students developing long-term career aspirations in the field of of motivational design lead to satisfying learning experiences. When
AI. Therefore, we considered the relationships between six constructs appropriately designed and implemented, the ARCS model’s four factors
(intrinsic motivation, career motivation, attention, relevance, confi- maintain students’ continuous interest and learning motivation (Keller,
dence, and satisfaction), and used a structural equation modeling (SEM) 2010). Past and recent studies have employed the ARCS model to
to examine the predictive relations among in an AI curriculum imple- examine the motivational effects of various instructional designs, espe-
mented in selected Beijing primary schools. There are two research cially in emerging technology-integrated learning environments (e.g.,
questions guided the current research: (1) Are the hypothesized re- STEM learning, Julia and Antolí, 2019; ubiquitous learning, Chen and
lationships in the motivational model supported? (2) Do any gender Chen, 2018; virtual reality, Villena Taranilla, C ozar-Gutierrez,
differences influence the measured variables? Gonzalez-Calero and L opez Cirugeda, 2019; AR-based ubiquitous
learning systems, Huang et al., 2019). These studies have all highlighted
2. Literature review the role of attention, relevance, and confidence strategies play in
improving students’ learning satisfaction and maintaining their motiva-
2.1. Motivation for AI learning tion when the ARCS model is properly integrated into the lesson design.
Therefore, this study attempted to foster students’ motivation to learn
Research in recent decades has strongly evidenced the role motiva- AI through ARCS motivational strategies by attracting and sustaining
tion plays in enhancing students’ willingness to learn (e.g., Chai et al., students’ attention to the learning content (attention) (Hung et al.,
2020; Jong, 2014; Jong et al., 2006; Lan et al., 2018; Pintrich, 2003). As 2013); comprehending and establishing a relationship between the
highlighted by Ryan and Deci (2000) and Elliott and Dweck (1988), content and the students (relevance) (Keller and Suzuki, 2004); facili-
intrinsic motivation and career motivation can universally explain how tating positive learning experiences and building confidence in a
people’s self-motivated beliefs direct their goal-oriented behaviors. technology-based environment (confidence) (Hong et al., 2017); and
People feel intrinsically motivated when they are interested in and enjoy promoting students’ satisfaction with the learning procedure or outcome
a learning activity and are satisfied by it (Ryan and Deci, 2000). Studies (satisfaction) (Keller, 2010).
have highlighted that intrinsic motivation directs academic goal-setting Previous studies have conducted surveys to validate the ARCS model
and sustains students’ engagement in relevant activities, which in turn and have explored the interaction among its constituting factors in
results in the achievement of intended learning outcomes and career different educational settings (e.g., a college-level course, Huang et al.,
plans (Froiland and Worrell, 2016; Jong et al., 2010; Lauermann et al., 2006; education for senior citizens, Loorbach et al., 2015). Their findings
2017). indicate the effectiveness of motivational strategies in enhancing stu-
Career motivation stems from students engaging in tasks and activ- dents’ motivation. However, the ARCS model does not seem to have been
ities of a particular subject when they perceive high value of a future employed to examine primary students’ motivation to learn AI through
career and successful performance in that subject (Eccles and Wigfield, SEM. Therefore, a contextualized study with the ARCS motivational
2020; Glynn et al., 2009). For example, students interested in pursuing strategies was designed to validate primary students’ motivation for
STEM-related careers have been found to be highly motivated to learn learning AI.
science upon recognizing its utilitarian value (Rozek et al., 2015).
Conversely, students disinterested in science-related careers were poorly 2.3. Gender differences
motivated to learn STEM subjects (Rosenzweig and Wigfield, 2016).
Therefore, educators who wish to cultivate a considerably large number Gender differences in technology-based learning have emerged as an
of future AI engineers must develop strategies to cultivate students’ issue that merits investigation (Cai et al., 2017). Research has provided
career motivation toward the field of AI. To this end, building students’ different explanations for STEM gender gap in students’ motivated
initial interest in AI through motivation-based instructional design may achievement-related choices (e.g., Lee et al., 2019; Wang et al., 2015;
be beneficial. Wang and Degol, 2017), including gender differences in psychological
Hidi and Harackiewicz (2000) argued that interests and goals can processes (e.g., motivations, Degol et al., 2018), social-cognitive career
cultivate both intrinsic and extrinsic motivation in people. Accordingly, theory (e.g., Lent et al., 2018), developmental processes (e.g.,
this study assumed that in the initial stages, students learn AI because self-theories, Fisher et al., 2020). The current research focused on the

2
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

psychological process (i.e., motivations) for improving students’ career H10. Confidence predicts satisfaction.
motivation.
H11. Confidence predicts CM.
Research has indicated male students had higher intrinsic motivation
in pursuing STEM career in future (e.g., Degol et al., 2018; Moe et al., H12. Relevance predicts satisfaction.
2018). In addition, incorporating AI technology in a course implies new
H13. Relevance predicts CM.
instructions. Students perceived teachers’ instructional efforts to
improve and maintain students motivation could impact how students H14. Satisfaction predicts CM.
motivation change during the instruction (Cheng et al., 2020; Young
et al., 2018). Notable gender gaps in STEM courses have been observed in 3. Method
mathematics and the physical sciences and related to students
self-confidence (e.g., Litzler et al., 2014), in programming experience and 3.1. Participants
related to students STEM motivation (e.g., Master et al., 2017).
AI is a sub-discipline of computer engineering. The design of AI ap- Through convenience sampling, 420 students (male, 50.7%; female,
plications depends on programming, scientific thinking, and mathemat- 49.3%) were recruited from primary schools in Beijing, China. The par-
ical algorithms, which explains why male and female students may react ticipants comprised fourth to sixth graders, with an average age of 10.23
differently to learning AI. Hence, an empirical examination of the po- (SD ¼ 0.75) years. All recruited students took an AI course, and the mean
tential gender differences that may underlie students’ motivational fac- time spent by them on AI-related activities was 6.69 (SD ¼ 5.32) hours.
tors is necessary. At the end of their AI course, they were invited by their teachers to
voluntarily participate in a 15-min online survey.
2.4. Purpose of the present study
3.2. The instructional context
As suggested by the literature review, motivational factors are
multifaceted, and an examination of these motivational factors in the The AI curriculum taught in the selected schools was based on a new
context of AI learning in primary schools is essential for constructing AI textbook (Qin et al., 2019) issued by the local education authority.
curricula with a motivational design. Students with high intrinsic moti- Through class observations and textbook content, teachers taught the AI
vation usually show higher engagement during learning and achieve- course through direct instruction followed by hands-on activities related
ment, that predicts career orentation (Froiland and Worrell, 2016; Ryan to the knowledge and applications taught. Through this course, students
and Deci, 2000). Within the ARCS model, Keller (2010) explains that acquired basic AI knowledge, including the meaning of data, data
instructional strategies that promotes student attention, perceptions of collection and processing, and the outputs generated, and learned about
content relevance, and confidence, are important predictors of satisfac- AI applications, including facial and voice recognition, content recom-
tion. Therefore, this study proposed a model to explain the relationships mendation, and machine learning. A device similar to Micro:bit with a
between intrinsic motivation, career motivation, and the ARCS motiva- Chinese coding environment supported students’ AI learning.
tional components in a newly designed AI curriculum. Accordingly, our Supported by classroom observations, a teaching design based on the
model examined the hypotheses for the SEM proposed below: ARCS motivational components was implemented in the teaching pro-
Fig. 1 illustrates the study’s theoretical model. The hypotheses (H) cess. The motivational design strategies focused on the following: (1)
tested in the model include the following: attention: AI-related activities that demonstrate the usefulness of AI such
as speedy object recognition; students experienced using AI applications
H1. Intrinsic motivation (IM) predicts career motivation (CM). such as a poetry generator; (2) relevance: students were asked to recall
H2. IM predicts attention. personal experiences related to the content being learned; AI applications
that help them complete inquiry tasks, such as flower recognition apps;
H3. IM predicts confidence. (3) confidence: the teacher encouraged the students to share their
H4. IM predicts relevance. learning experiences to achieve the course’s learning goals and gave
encouraging feedback; (4) satisfaction: the teacher motivated the stu-
H5. Attention predicts confidence. dents through verbal praises and encouraging feedback. Generally, the
H6. Attention predicts relevance. lessons were well paced with sufficient challenging coding activities that
were conducted under the guidance of teachers.
H7. Attention predicts satisfaction.
H8. Attention predicts CM. 3.3. Instrument

H9. Confidence predicts relevance. An instrument was designed to examine students’ motivation to learn
AI using a 4-point Likert scale (1 ¼ strongly disagree; 4 ¼ strongly agree).
Additionally, the survey included items regarding the participating stu-
dents’ demographic information, namely gender, age, grade level, and
hours spent on AI learning and projects. The instrument items were
adopted from several sources: (1) the three items of intrinsic motivation
were adapted from the Motivated Strategies for Learning Questionnaire
(MSLQ) (Pintrich et al., 1991) (e.g., In the AI class, I prefer lesson ma-
terials that arouse my curiosity, even if they are difficult to learn); (2) for
career motivation (three items), two items were adapted from Glynn
et al. (2009) and one was adapted from Vennix, den Brok, and Taconis
(2018) (e.g., I think that learning AI can help my future career); (3)
attention was examined through three self-constructed items (e.g., I
actively pay attention to AI applications); (4) the four items of relevance
were adapted from Keller (2010) (e.g., I am aware that AI technology will
change the world); (5) confidence was assessed through two items
Fig. 1. The theoretical model proposed in this paper. modified from Keller (2010) and one item modified from the MSLQ

3
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

(Pintrich et al., 1991) (e.g., I am taking this AI class because I believe that Table 2
I can succeed if I try hard enough); and (6) the three items for satisfaction Standardized and unstandardized coefficients of CFA.
were modified from Keller (2010) (e.g., I feel satisfied with what I am Observed Standardized Unstandardized Standard Critical
gaining from the AI class). The finalized instrument was reviewed by four variable estimate estimate error ratio (p-
researchers who specialize in the fields of curriculum and instruction and value)
psychology, and 12 primary school teachers of the AI curriculum. Table 1 Intrinsic motivation
presents the means, standard deviations, and Cronbach’s alphas of the 1 0.88 1.00
composite variables. 2 0.84 0.98 0.04 22.03***
3 0.81 0.95 0.05 20.88***
Career motivation
1 0.84 1.00
3.4. Data analysis 2 0.76 0.96 0.06 17.45***
3 0.80 1.01 0.05 18.61***
Prior to the analyses, tests of univariate and multivariate normality Attention
were conducted to ensure that the model was acceptable. Subsequently, 1 0.90 1.00
2 0.86 0.98 0.04 24.29***
the data were analyzed guided by the study’s research questions. 3 0.60 0.73 0.05 13.73***
Univariate normality was tested in accordance with Kline’s (2005) Relevance
criteria. The values of skewness (ranging from 1.65 to 1.00) and 1 0.67 1.00
kurtosis (ranging from 0.68 to 3.15) (see Table 1) indicated that the 2 0.86 1.16 0.08 15.27***
3 0.84 1.13 0.08 14.89***
dataset was normally distributed following the recommended value of
4 0.76 1.21 0.09 13.79***
under |3| and |10|, respectively. Mardia’s coefficient was used to Confidence
examine the multivariate normality of the measurement model (Raykov 1 0.82 1.00
and Marcoulides, 2008). Mardia’s coefficient obtained from the data was 2 0.83 0.86 0.04 19.45***
329.651, which was less than the recommended value of p (pþ2) ¼ 19 3 0.89 1.01 0.05 21.01***
Satisfaction
(19 þ 2) ¼ 399 (where p is the number of observed variables). Hence, the 1 0.82 1.00
model’s multivariate normality was supported. 2 0.88 0.99 0.05 21.30***
For the first research question, AMOS (version 23.0) was employed 3 0.85 0.99 0.05 20.26***
using maximum likelihood estimation to test whether the measurement Note. *** p < .001.
model’s fitness indices χ2/df (<5.0), GFI(>0.90), TLI(>0.90),
CFI(>0.90), RMSEA(<0.08), SRMR(<0.05) were acceptable. Thereafter,
demonstrated to influence the examined constructs, we decided to
SEM was used to examine the hypothetical model.
include them in the SEM analyses. As noted in Table 3, each subscale’s
For the second research question, instead of using a t-test and analysis
convergent validity was satisfied by the following criteria: the composite
of variance (ANOVA) to make cross-group comparisons, we followed
reliability (CR) (>0.7) and the average variance extracted (AVE) (>0.5)
Hong et al.’s (2003) recommendation to use the latent mean analysis
of each sub-scale (Hair et al., 2010). The discriminant validity indices
approach for latent variables that cannot be directly measured. Mea-
(see Table 3) were computed on the basis of the AVE values, and they
surement invariance (i.e., configural, metric, and scalar invariance tests)
confirmed that the discriminant validity was satisfied (Fornell and
is a prerequisite for making comparisons across groups using latent mean
Larcker, 1981).
analysis (Hong et al., 2003). Finally, measurement invariance tests were
conducted to determine the latent mean differences across groups
(Cheung and Rensvold, 2002). 4.2. Test of the structural model

4. Results After confirming the fit of the measurement model, the full model
with hypothesized relationships was statistically tested to perform a path
4.1. Test of the measurement model analysis using maximum likelihood estimation. The hypothesized model
had good fit: χ2/df ¼ 2.170, GFI ¼ 0.930, TLI ¼ 0.966, CFI ¼ 0.973,
Confirmatory factor analysis (CFA) was conducted to test the mea- RMSEA ¼ 0.053, SRMR ¼ 0.032 (Hair et al., 2010). The standardized
surement model. The overall model fit was assessed using six goodness- path coefficients for the research model are presented in Fig. 2 and
of-fit indices: χ2/df ¼ 2.186, GFI ¼ 0.930, TLI ¼ 0.966, CFI ¼ 0.973, Table 4. The results indicated that people’s intrinsic motivation is the
RMSEA ¼ 0.053, SRMR ¼ 0.032 (Hair et al., 2010). The results confirmed strongest predictor of career motivation (H1; β ¼ 0.37, p < .001).
that the survey items had satisfactory construct validity in the AI learning Additionally, intrinsic motivation was found to predict attention (H2; β
context. Table 2 presents the standardized and unstandardized parameter ¼ 0.86, p < .001), confidence (H3; β ¼ 0.46, p < .001), and relevance
estimates. All item parameters are statistically significant at the p < .001 (H4; β ¼ 0.33, p < .001). Attention improved confidence (H5; β ¼ 0.37, p
level. < .001), and was related to relevance (H6; β ¼ 0.36, p < .001), and
Pearson’s correlations were computed to investigate the relationships promoted career motivation (H8; β ¼ 0.25, p < .05); however, attention
among the motivational factors to serve as a foundation for SEM. The did not predict satisfaction directly (H7; β ¼ 0.11, p > .05). In our study,
students’ perceptions of motivational factors were significantly corre- confidence directly predicted relevance (H9; β ¼ 0.22, p < .01), satis-
lated (r ¼ 0.58–0.75; p < .001; see Table 3). Given the variables were faction (H10; β ¼ 0.24, p < .001), and career motivation (H11; β ¼ 0.17,
p < .05). Relevance predicted satisfaction (H12; β ¼ 0.56, p < .001) and
career motivation (H13; β ¼ 0.26, p < .01). Satisfaction did not predict
Table 1
Descriptive statistics of measured variables.
career motivation directly (H14; β ¼ 0.11, p > .05).

Variable Mean SD α Skewness Kurtosis


4.3. Invariance tests
Intrinsic motivation 3.33 0.73 0.88 1.07 0.78
Career motivation 3.30 0.72 0.85 1.06 1.03
Attention 3.38 0.68 0.82 1.00 0.68 Several invariance tests were performed before conducting the latent
Relevance 3.42 0.65 0.86 1.33 1.95 mean analysis. First, configural invariance was assessed without con-
Confidence 3.41 0.68 0.88 1.24 1.55 straining equality across the groups (i.e., male and female). The results of
Satisfaction 3.46 0.67 0.89 1.65 3.15
goodness-of-fit (χ2 ¼ 617.496, df ¼ 274, χ2/df ¼ 2.254, CFI ¼ 0.943,

4
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

Table 3
Correlation analyses of the relationships among the measured variables.
CR AVE 1 2 3 4 5 6

1. Intrinsic motivation 0.88 0.71 (0.84)


2. Career motivation 0.84 0.64 0.70 (0.8)
3. Attention 0.84 0.64 0.67 0.66 (0.8)
4. Relevance 0.86 0.62 0.63 0.73 0.65 (0.79)
5. Confidence 0.88 0.71 0.75 0.70 0.69 0.65 (0.84)
6. Satisfaction 0.89 0.73 0.70 0.67 0.65 0.58 0.72 (0.85)

Note. All correlations were statistically significant at p < .001.


Diagonal elements are the square roots of the AVE; off-diagonal elements are the correlation estimates.

4.4. Latent mean analysis

The latent mean analysis was performed by constraining the male


group’s latent means to be zero. This was done to use the male group as a
reference group for comparison with the female group. The latent mean
represented a good model fit (χ2 ¼ 668.849, df ¼ 300, χ2/df ¼ 2.229,
CFI ¼ 0.939, RMSEA ¼ 0.054). Table 6 presents the results of the latent
variables with significant gender differences for the factors of intrinsic
motivation, career motivation, attention, relevance, confidence, and
satisfaction. Overall, male students had higher mean values for all factors
than female students did.

Table 5
Summary of invariance tests.
Fig. 2. Structural model of the measured factors. Note. * p < .05; ** p < .01; ***
p < .001. X2 Df X2/df p- CFI △CFI RMSEA
value

RMSEA ¼ 0.055) in the configural invariance test (see Table 5) indicated Configural 617.496 274 2.254 .000 0.943 0.055
invariance
that the structural patterns were similar across groups. This implied that
Metric 651.887 287 2.271 .000 0.939 0.004 0.055
the configural model could serve as a baseline for comparison with other invariance
restricted models in the invariance tests. Subsequently, metric invariance Scalar 687.404 306 2.246 .000 0.937 0.002 0.055
was assessed by constraining the factor loadings to be equal across invariance
groups. The results demonstrated a good model fit: χ2 ¼ 651.887, df ¼
287, χ2/df ¼ 2.271, CFI ¼ 0.939, RMSEA ¼ 0.055. Finally, a scalar
invariance test was performed by constraining the intercepts across Table 6
groups to be invariant. The model fit indices showed that χ2 ¼ 687.404, Tests of latent mean differences.
df ¼ 306, χ2/df ¼ 2.246, CFI ¼ 0.937, RMSEA ¼ 0.055. Because the χ2 Differences in latent mean C.R.
test is highly sensitive to sample sizes (Quintana and Maxwell, 1999), we Intrinsic motivation 0.30 4.17***
adopted the criteria recommended by Cheung and Rensvold (2002). The Career motivation 0.20 2.76**
absolute value of △CFI was <0.01, suggesting that the invariance hy- Attention 0.24 3.43***
Relevance 0.14 2.45*
potheses were supported. Given that the assumptions of configural,
Confidence 0.22 3.14**
metric, and scalar invariance were satisfied, a latent mean analysis could Satisfaction 0.16 2.43*
be conducted.
Note. * p < .05; ** p < .01; *** p < .001.

Table 4
Hypothesis testing through SEM.
Hypothesis From To Unstandardized estimate Standardized estimate Critical ratio (p-value) Hypotheses supported?

H1 Intrinsic motivation Career motivation 0.34 0.37 3.63*** Yes


H2 Intrinsic motivation Attention 0.84 0.86 19.42*** Yes
H3 Intrinsic motivation Confidence 0.44 0.46 4.68*** Yes
H4 Intrinsic motivation Relevance 0.25 0.33 3.37*** Yes
H5 Attention Confidence 0.36 0.37 3.82*** Yes
H6 Attention Relevance 0.28 0.36 3.76*** Yes
H7 Attention Satisfaction 0.10 0.11 1.34 No
H8 Attention Career motivation 0.24 0.25 2.56* Yes
H9 Confidence Relevance 0.18 0.22 3.15** Yes
H10 Confidence Satisfaction 0.23 0.24 3.49*** Yes
H11 Confidence Career motivation 0.18 0.17 2.40* Yes
H12 Relevance Satisfaction 0.66 0.56 6.55*** Yes
H13 Relevance Career motivation 0.32 0.26 2.65** Yes
H14 Satisfaction Career motivation 0.12 0.11 1.43 No

Note. * p < .05; ** p < .01; *** p < .001.

5
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

5. Discussion and conclusions career orientation (Degol et al., 2018; Wang et al., 2015). The responses
from the male and female students in this study indicated that female
Given that students will work in an AI-empowered society in the students may perceive themselves as being less motivated to learn AI.
future, fueling students’ motivation to learn AI is crucial. This study This suggests that facilitating female students’ motivation by cultivating
tested an SEM of motivational factors, comprising intrinsic motivation, their interests in AI through proper instructional strategies is critical. For
career motivation, and the ARCS motivational strategies (i.e., attention, example, teachers may highlight some very successful female AI engi-
relevance, confidence, and satisfaction) among primary students (Grades neers who have contributed to the field to inspire female students.
4–6) in Beijing, China who were attending AI lessons. Accordingly, the In conclusion, this study investigated the motivational factors and
following two research objectives were executed: (1) testing a hypo- strategies that influenced students’ AI learning and discovered that
thetical model with six motivational factors and strategies: intrinsic increasing students’ intrinsic motivation, sustaining their attention to AI,
motivation, career motivation, attention, relevance, confidence, and highlighting the relevance of AI, and promoting confidence in learning AI
satisfaction; and (2) examining the potential effect of gender differences results in enhanced career motivation. Additionally, research has indi-
on these motivational factors and strategies that influence male and fe- cated gender effects on the pursuit of certain career paths (STEM; Wang
male students’ motivation to learn AI. et al., 2015). This study tested gender group differences in the AI learning
For the first research objective of testing the hypotheses, the results motivational model and discovered that female students perceived
indicated that 12 out of 14 hypotheses were supported. Consistent with themselves to be less motivated to learn AI. Overall, the results suggest
previous studies, this model highlighted that intrinsic motivation (e.g., that educators should foster students’ intrinsic motivation and design
interests) is the most important factor that directly predicts students’ appropriate instructional strategies to motivate students during the AI
career orientation (H1) (Degol et al., 2018; Ryan and Deci, 2000) and learning process. This will strengthen students’ career motivation to
motivational strategies, including attention (H8), confidence (H11), and pursue AI-related knowledge and careers.
relevance (H13) (Su and Cheng, 2015). According to the hypothesized
motivational model, SEM results confirmed that intrinsic motivation also
Declaration of competing interest
predicted attention (H2), confidence (H3), and relevance (H4). In addi-
tion, the correlation analysis showed that attention, confidence, rele-
Research approval was obtained from the Survey and Behavioural
vance and satisfaction were positively and significantly correlated. When
Research Ethics Committee in the authors’ institution.
analyzing results, it is observed that attention is a predictive of confi-
The authors have no conflict of interest.
dence (H5) and relevance (H6). Confidence is a predictive of relevance of
AI (H9). Also, confidence (H10), and relevance (H12) were significant
Acknowledgements
predictors for satisfaction This is aligned with previous studies that
examine students’ motivation using the ARCS model (Huang et al., 2006;
We appreciate the support offered by the school principals to the
Loorbach et al., 2015). The findings attest to the quality of the designed
study. We are also thankful to the teachers and students for participating
AI curriculum and implementation. Overall, the model depicts three
in the experiment.
interrelated forms of motivation that AI curriculum designers may need
to address. The intrinsic motivation that students possess before they
References
learn about AI; the instructional motivation their experience as a result of
the designed curriculum; and the career motivation which represent the Cai, Z., Fan, X., & Du, J. (2017). Gender and attitudes toward technology use: a meta-
continuous intention to learn about AI after the course. The model we analysis. Computers & Education, 105, 1–13.
obtained is generally consistent with recent study that emphasizes Chai, C. S., Wang, X., & Xu, C. (2020). An extended theory of planned behavior for the
modelling of Chinese secondary school students’ intention to learn Artificial
enhancing students’ intention to learn about AI by highlighting the
Intelligence. Mathematics, 8(11), 2089. https://doi.org/10.3390/math8112089
relevance of AI through helping students to appreciate the usefulness of Chen, C. C., & Chen, C. Y. (2018). Exploring the effect of learning styles on learning
AI and its potential to promote social good (Chai et al., 2020). The study achievement in a u-Museum. Interactive Learning Environments, 26(5), 664–681.
Chen, X., Xie, H., Zou, D., & Hwang, G. J. (2020). Application and theory gaps during the
informs educators about the possibility of engaging students to learn AI
rise of artificial intelligence in education. Computers and Education: Artificial
in primary school with good motivational effects. Moreover, the results Intelligence, 1, 100002.
corresponded to that of previous studies that have suggested that the Cheng, L., Antonenko, P. D., Ritzhaupt, A. D., Dawson, K., Miller, D., MacFadden, B. J.,
ARCS motivational model could provide instructors with a means to meet Grant, C., Sheppard, T. D., & Ziegler, M. (2020). Exploring the influence of teachers’
beliefs and 3D printing integrated STEM instruction on students’ STEM motivation.
students’ motivational demands (Ucar and Kumtepe, 2020; Wah, 2015). Computers & Education, 158, 103983.
Nonetheless, attention did not directly predict satisfaction (H7), and Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing
satisfaction did not directly predict career motivation (H14). This indi- measurement invariance. Structural equation modeling, 9(2), 233–255.
Degol, J. L., Wang, M. T., Zhang, Y., & Allerton, J. (2018). Do growth mindsets in math
cated that attention alone cannot satisfy students’ AI learning needs. We benefit females? Identifying pathways between gender, mindset, and motivation.
found that within the ARCS model, the influence of attention is mediated Journal of youth and adolescence, 47(5), 976–990.
by relevance and confidence. Apparently, relevance of AI and confidence Eccles, J. S., & Wigfield, A. (2020). From expectancy-value theory to situated expectancy-
value theory: a developmental, social cognitive, and sociocultural perspective on
in AI may constitute key elements to focus the students’ attention on AI motivation. Contemporary Educational Psychology, 101859.
and thus increase their satisfaction. The findings reflect that reinforcing Elliott, E. S., & Dweck, C. S. (1988). Goals: an approach to motivation and achievement.
the relevance of AI and confidence in AI may enhance satisfaction. Journal of Personality and Social Psychology, 54(1), 5–12.
Fisher, C. R., Thompson, C. D., & Brookes, R. H. (2020). Gender differences in the Australian
The findings provide insights into the motivational design of AI
undergraduate STEM student experience: a systematic review (pp. 1–14). Higher
curricula. From introducing students to powerful AI apps that help them Education Research & Development.
relate the subject matter to their personal life experiences to unpacking Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with
unobservable and measurement error. Journal of Marketing Research, 18, 39–50.
complex knowledge pertaining to AI, different methods can be adopted to
Froiland, J. M., & Worrell, F. C. (2016). Intrinsic motivation, learning goals, engagement,
construct an effective motivational design. This seems to correspond to and achievement in a diverse high school. Psychology in the Schools, 53(3), 321–336.
the authentic learning process in general (Wah, 2015). Fu, S., Gu, H., & Yang, B. (2020). The affordances of AI-enabled automatic scoring
For the second research objective concerning gender differences in applications on learners’ continuous learning intention: an empirical study in China.
British Journal of Educational Technology. https://doi.org/10.1111/bjet.12995
the motivational factors and strategies in the AI learning context, the Geng, J., Jong, M. S. Y., & Chai, C. S. (2019). Hong Kong teachers’ self-efficacy and
results revealed statistical gender differences in terms of intrinsic moti- concerns about STEM education. The Asia-Pacific Education Researcher, 28(1), 35–45.
vation, career motivation, attention, relevance, confidence, and satis- Glynn, S. M., Taasoobshirazi, G., & Brickman, P. (2009). Science motivation
questionnaire: construct validation with nonscience majors. Journal of Research in
faction. This finding is consistent with that of previous studies that have Science Teaching: The Official Journal of the National Association for Research in Science
reported a significant gender difference in STEM-learning motivation and Teaching, 46(2), 127–146.

6
P.-Y. Lin et al. Computers and Education: Artificial Intelligence 2 (2021) 100006

Hair, J. F., Jr., Black, W. C., Babin, B. J., Anderson, R. E., & Tatham, R. L. (2010). SEM: an Master, A., Cheryan, S., Moscatelli, A., & Meltzoff, A. N. (2017). Programming experience
Introduction. Multivariate Data Analysis: A Global Perspective (seventh ed., pp. promotes higher STEM motivation among first-grade girls. Journal of experimental
629–686). Upper Saddle River, NJ: Pearson Education. child psychology, 160, 92–106.
Hidi, S., & Harackiewicz, J. M. (2000). Motivating the academically unmotivated: a Moe, A., Jansen, P., & Pietsch, S. (2018). Childhood preference for spatial toys. Gender
critical issue for the 21st century. Review of educational research, 70(2), 151–179. differences and relationships with mental rotation in STEM and non-STEM students.
Hong, J. C., Hwang, M. Y., Tai, K. H., & Tsai, C. R. (2017). An exploration of students’ Learning and Individual Differences, 68, 108–115.
science learning interest related to their cognitive anxiety, cognitive load, self- Pintrich, P. R. (2003). A motivational science perspective on the role of student
confidence and learning progress using inquiry-based learning with an iPad. Research motivation in learning and teaching contexts. Journal of Educational Psychology,
in Science Education, 47(6), 1193–1212. 95(4), 667–686.
Hong, S., Malik, M. L., & Lee, M. K. (2003). Testing configural, metric, scalar, and latent Pintrich, P. R., Smith, D. A. F., Garcia, T., & McKeachie, W. J. (1991). A Manual for the Use
mean invariance across genders in sociotropy and autonomy using a non-Western of the Motivated Strategies for Learning. Michigan: School of Education Building, The
sample. Educational and psychological measurement, 63(4), 636–654. University of Michigan.
Huang, T. C., Chen, M. Y., & Hsu, W. P. (2019). Do learning styles matter? Motivating Qin, J. J., Ma, F. G., & Guo, Y. M. (2019). Foundations of Artificial Intelligence for Primary
learners in an Augmented geopark. Journal of Educational Technology & Society, 22(1), School. Beijing, CN: Popular Science Press.
70–81. Quintana, S. M., & Maxwell, S. E. (1999). Implications of recent developments in
Huang, W., Huang, W., Diefes-Dux, H., & Imbrie, P. K. (2006). A preliminary validation of structural equation modeling for counseling psychology. The Counseling Psychologist,
Attention, Relevance, Confidence and Satisfaction model-based Instructional Material 27(4), 485–527.
Motivational Survey in a computer-based tutorial setting. British Journal of Raykov, T., & Marcoulides, G. A. (2008). An Introduction to Applied Multivariate Analysis.
Educational Technology, 37(2), 243–259. Routledge.
Hung, I. C., Chao, K. J., Lee, L., & Chen, N. S. (2013). Designing a robot teaching assistant Rosenzweig, E. Q., & Wigfield, A. (2016). STEM motivation interventions for adolescents:
for enhancing and sustaining learning motivation. Interactive learning environments, a promising start, but further to go. Educational Psychologist, 51(2), 146–163.
21(2), 156–171. Rozek, C. S., Hyde, J. S., Svoboda, R. C., Hulleman, C. S., & Harackiewicz, J. M. (2015).
Hwang, G. J., Xie, H., Wah, B. W., & Gasevic, D. (2020). Vision, challenges, roles and Gender differences in the effects of a utility-value intervention to help parents
research issues of Artificial Intelligence in Education. Computers and Education: motivate adolescents in mathematics and science. Journal of Educational Psychology,
Artificial Intelligence, 1, 100001. 107(1), 195–206.
Jong, M. S. Y. (2014). Elementary students’ view of collaborative knowledge building in Ryan, R. M., & Deci, E. L. (2000). Intrinsic and extrinsic motivations: classic definitions
LearningVillages. Interaction Design and Architecture(s), 21, 37–56. and new directions. Contemporary educational psychology, 25(1), 54–67.
Jong, M. S. Y. (2020). Promoting elementary pupils’ learning motivation in Shang, J. J., Jong, M. S. Y., Lee, F. L., Lee, J. H. M., Wong, M., Luk, E., & Cheung, K.
environmental education with mobile inquiry-oriented ambience-aware fieldwork. (2006). Using the "Record-Replay" function for elaboration of knowledge in
International Journal of Environmental Research & Public Health, 17(7). Article 2504. educational games. Frontiers in Artificial Intelligence and Applications, 151, 503–506.
Jong, M. S. Y., Shang, J. J., Lee, F. L., Lee, J. H. M., & Law, H. Y. (2006). Learning online: a Scherrer, V., & Preckel, F. (2019). Development of motivational variables and self-esteem
comparative study of a game-based situated learning approach and a traditional web- during the school career: a meta-analysis of longitudinal studies. Review of
based learning approach. In Z. Pan, R. Aylett, H. Diener, X. Jin, S. Gobel, & L. Li Educational Research, 89(2), 211–258.
(Eds.), Lecture Notes in Computer Science: Technologies for E-Learning and Digital So, H. J., Jong, M. S. Y., & Liu, C. C. (2020). Computational thinking education in the
Entertainment (pp. 541–551). Heidelberg: Springer. Asian Pacific region. The Asia-Pacific Education Researcher, 29, 1–8.
Jong, M. S. Y., Shang, J. J., Lee, F. L., & Lee, J. H. M. (2010). Constructivist Learning through Starcic, A. I. (2019). Human learning and learning analytics in the age of artificial
Computer Gaming. Technologies Shaping Instruction and Distance Education: New Studies intelligence. British journal of educational technology, 50(6), 2974–2976.
and Utilization (pp. 207–222). New York: Information Science Reference. Su, C. H., & Cheng, C. H. (2015). A mobile gamification learning system for improving the
Juli
a, C., & Antolí, J.O. (2019). Impact of implementing a long-term STEM-based active learning motivation and achievements. Journal of Computer Assisted Learning, 31(3),
learning course on students’ motivation. International Journal of Technology and Design 268–286.
Education, 29(2), 303–327. Ucar, H., & Kumtepe, A. T. (2020). Effects of the ARCS-V-based motivational strategies on
Keller, J. M. (1987). Development and use of the ARCS model of instructional design. online learners’ academic performance, motivation, volition, and course interest.
Journal of Instructional Development, 10(3), 2–10. Journal of Computer Assisted Learning, 36(3), 335–349.
Keller, J. M. (2010). The Arcs model of motivational design. In Motivational Design for Veletsianos, G., & Russell, G. S. (2014). Pedagogical agents. In Handbook of Research on
Learning and Performance (pp. 43–74). Boston, MA: Springer. Educational Communications and Technology (pp. 759–769). New York, NY: Springer.
Keller, J., & Suzuki, K. (2004). Learner motivation and e-learning design: a Vennix, J., den Brok, P., & Taconis, R. (2018). Do outreach activities in secondary STEM
multinationally validated process. Journal of educational Media, 29(3), 229–239. education motivate students and improve their attitudes towards STEM? International
Kline, R. B. (2005). Principles and Practices of Structural Equation Modeling. New York: The Journal of Science Education, 40(11), 1263–1283.
Guildford Press. Villena Taranilla, R., C ozar-Gutierrez, R., Gonzalez-Calero, J. A., & L
opez Cirugeda, I.
Knox, J. (2020). Artificial Intelligence and Education in China. Learning, Media and (2019). Strolling through a city of the Roman Empire: an analysis of the potential of
Technology. https://doi.org/10.1080/17439884.2020.1754236 virtual reality to teach history in Primary Education. Interactive Learning Environments,
Lan, Y. J., Botha, A., Shang, J. J., & Jong, M. S. Y. (2018). Technology enhanced 1–11.
contextual game-based language learning. Educational Technology & Society, 21(3), Wah, L. K. (2015). The effects of instruction using the arcs model and geogebra on Upper
86–89. secondary students’ motivation and achievement in learning combined
Lauermann, F., Tsai, Y.-M., & Eccles, J. S. (2017). Math-related career aspirations and transformation. Asia Pacific Journal of Educators and Education, 30, 141–158.
choices within Eccles et al.’s expectancy–value theory of achievement-related Wang, M. T., & Degol, J. L. (2017). Gender gap in science, technology, engineering, and
behaviors. Developmental Psychology, 53(8), 1540–1559. mathematics (STEM): current knowledge, implications for practice, policy, and future
Lee, M. H., Chai, C. S., & Hong, H. Y. (2019). STEM education in Asia Pacific: challenges directions. Educational psychology review, 29(1), 119–140.
and development. The Asia-Pacific Education Researcher, 28, 1–4. Wang, M. T., Degol, J., & Ye, F. (2015). Math achievement is important, but task values
Lent, R. W., Sheu, H. B., Miller, M. J., Cusick, M. E., Penn, L. T., & Truong, N. N. (2018). are critical, too: examining the intellectual and motivational factors leading to gender
Predictors of science, technology, engineering, and mathematics choice options: a disparities in STEM careers. Frontiers in psychology, 6, 36.
meta-analytic path analysis of the social–cognitive choice model by gender and race/ Yannier, N., Hudson, S. E., & Koedinger, K. R. (2020). Active learning is about more than
ethnicity. Journal of Counseling Psychology, 65(1), 17–35. hands-on: a mixed-reality AI system to support STEM education. International Journal
Li, K., & Keller, J. M. (2018). Use of the ARCS model in education: a literature review. of Artificial Intelligence in Education, 1–23.
Computers & Education, 122, 54–62. Young, A. M., Wendel, P. J., Esson, J. M., & Plank, K. M. (2018). Motivational decline and
Litzler, E., Samuelson, C. C., & Lorah, J. A. (2014). Breaking it down: engineering student recovery in higher education STEM courses. International Journal of Science Education,
STEM confidence at the intersection of race/ethnicity and gender. Research in Higher 40(9), 1016–1033.
Education, 55, 810–832. Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review
Loorbach, N., Peters, O., Karreman, J., & Steehouder, M. (2015). Validation of the of research on artificial intelligence applications in higher education–where are the
Instructional Materials Motivation Survey (IMMS) in a self-directed instructional educators? International Journal of Educational Technology in Higher Education, 16(1),
setting aimed at working with technology. British journal of educational technology, 39.
46(1), 204–218.

You might also like