lOMoARcPSD| 59062190
1
2
Hindawi
Complexity
Volume 2021, Arcle ID 8812542, 18 pages hps://doi.org/10.1155/2021/8812542
Review Arcle
A Review of Artificial Intelligence (AI) in Education from
2010 to 2020
Xuesong Zhai , Xiaoyan Chu ,
1
Ching Sing Chai , Morris Siu Yung Jong ,
2
Andreja
Istenic ,
3,4,5
Michael Spector ,
6
Jia-Bao Liu ,
7
Jing Yuan,
8
and Yan Li
1
1
Zhejiang University, Hangzhou 310058, China 2
Chinese University of Hong Kong, Hong Kong 999077, Hong Kong 3
University of Primorska, Faculty of Educaon, Koper 6000, Slovenia
4 University of Ljubljana, Faculty of Civil and Geodec Engineering, Ljubljana 1000, Slovenia 5
Federal University of Kazan, Instute of Psychology and Educaon, Kazan 420008, Russia 6
University of North Texas, Denton 76207, USA
7
Anhui Jianzhu University, Hefei 230601, China 8
Anhui Xinhua University, Hefei 230088, China
Correspondence should be addressed to Yan Li; yanli@zju.edu.cn
Received 27 August 2020; Revised 18 January 2021; Accepted 2 April 2021; Published 20 April 2021
Academic Editor: Ning Cai
Copyright © 2021 Xuesong Zhai et al. This is an open access arcle distributed under the Creave Commons Aribuon License,
which permits unrestricted use, distribuon, and reproducon in any medium, provided the original work is properly cited.
This study provided a content analysis of studies aiming to disclose how arficial intelligence (AI) has been applied to the
educaon sector and explore the potenal research trends and challenges of AI in educaon. A total of 100 papers including 63
empirical papers (74 studies) and 37 analyc papers were selected from the educaon and educaonal research category of
Social Sciences Citaon Index database from 2010 to 2020. The content analysis showed that the research quesons could be
classified into development layer (classificaon, matching, recommendaon, and deep learning), applicaon layer (feedback,
reasoning, and adapve learning), and integraon layer (affecon compung, role-playing, immersive learning, and
gamificaon). Moreover, four research trends, including Internet of Things, swarm intelligence, deep learning, and neuroscience,
as well as an assessment of AI in educaon, were suggested for further invesgaon. However, we also proposed the challenges
in educaon may be caused by AI with regard to inappropriate use of AI techniques, changing roles of teachers and students, as
well as social and ethical issues. The results provide insights into an overview of the AI used for educaon domain, which helps
to strengthen the theorecal foundaon of AI in educaon and provides a promising channel for educators and AI engineers to
carry out further collaborave research.
1. Introduction
The emergence of big data, cloud compung, arficial
neural networks, and machine learning has enabled
engineers to create a machine that can simulate human
intelligence. Building on these technologies, this study
refers to machines that are able to perceive, recognize,
learn, react, and solve problems as arficial intelligence (AI)
[1, 2]. Inevitably, such smart technologies will revoluonize
the workplaces of the future [3]. Thus, while AI can interact
and help humans perform at higher levels, it is emerging as
the next disrupve innovaon [4]. AI is currently viewed by
many as a driver that is integral to the fourth industrial
revoluon, and it may trigger the fourth revoluon in
educaon. Learning about AI has also begun to be part of
school curriculum [5, 6]. However, just as the emergence of
television and computers was once touted to be game
changers of educaon, they have been shown to in fact
enhance access to informaon without substanally
changing the core educaonal pracces. Nonetheless,
educators are obliged to review current AI capabilies and
idenfy possible pathways to opmize learning. Given the
lOMoARcPSD| 59062190
2 Complexity
increasing aenon, it is mely to review recent AI research
in educaon to provide educators with an updated
understanding of the field as a preparaon to possible
changes.
AI has been increasingly propagated as having strategic
value for educaon [7]. Loeckx [8] suggested that AI could
be an effecve learning tool that lessens the burdens of
both teachers and students and offers effecve learning
experiences for students. Coupled with current educaon
reforms such as the digitalizaon of educaonal resources,
gamificaon, and personalized learning experiences, there
are many opportunies for the development of AI
applicaons in educaon. For example, the modelling
potenal of AI techniques has been exploited systemacally
to develop reacve and adapve tutorials for the
construcon of individualized learning environments as
compensaon for the shortage of teachers through the use
of intelligent tutoring system (ITS) [10]. ITSs provide
personalized learning experience in four main ways:
monitoring students input, delivering appropriate tasks,
providing effecve feedback, and applying interfaces for
human-computer communicaon [7]. When more ITSs are
created for more subjects and topics, it is likely to change
the role of teachers, and hence, schooling may need to be
reconceptualized. There exist many concerns and worries
among teachers on if AI challenges their jobs. At the same
me, such quesons as what is being learned and how AI is
being used are being discussed currently by researchers as
well as by educaonal praconers. Some researchers
wondered whether advancements in AI would challenge or
even replace teachers since many other jobs are being
replaced by automaon [11]. There is an emerging
recognion that teachers’ professional roles need to be
adjusted as AI advances and this will trigger new
organizaonal forms [12]. Emerging challenges also
included students’ atudes towards these changes [13]. To
some extent, students as digital cizens are able to leverage
AI to improve learning outcomes. Nonetheless, they may fail
to use suitable AI techniques appropriately for a specific
learning context, which would result in negave atudes
towards learning [14].
To summarize, this research involves a review of the
studies of AI in educaon. Previous studies have included
three essenal perspecves of AI in knowledge processing:
(a) knowledge representaon, (b) knowledge obtaining, and
(c) knowledge derivaon [3]; this review will focus on AI
techniques and tools that have been integrated into
educaon recently aer the proliferaon of AI. The “first
generaon” of AI could support human intellectual work by
applying rule-based expert knowledge, and the “second
generaon” may find the opmal soluon by stascal/
search model, while the “third generaon” will dramacally
improve recognion performance based on the brain
model. This review focuses on arcles published in the
period from 2010 to 2020 from the Web of Science, as that
represents the period when the second and third generaon
of AI began to make headways into educaon. The research
quesons that guided this review are as follows:
(1) What is the overall state of AI in educaon? Which
research topics and research designs related to AI in
educaon are evident from 2010 to 2020?
(2) What are the trends in published studies in terms of
AI in educaon?
(3) What are the challenges generated from the current
research of AI in educaon?
2. Method
This study is a systemac literature review. The objecves of
the review were to analyze and interpret findings based on
predefined research quesons (see above) and criteria
which serve to point out future direcons [15]. The
predefined research foci as shown in Table 1 are research
purpose, learning subject, educaonal level, research
approach, and effects. The review was conducted in three
stages: planning, performing, and reporng the systemac
review.
2.1. Planning the Review. As previous reviews about AI were
conducted in the physical sciences [16, 17], the study aimed
to conduct a review in the field of the social sciences.
The Web of Science database and the Social Science
Citaon Index (SSCI) journals were selected for the search
for desired arcles published from 2010 to 2020. Arcles
published in the SSCI database are generally considered as
high-quality publicaon among educaon researchers. The
keyword employed was “arficial intelligence, and the
subject area was refined to educaon and educaonal
research”. This process yielded 142 arcles including 121
research arcles, 10 review papers, one interview paper,
and 5 book reviews. The selected arcles include both
analyc studies (primarily qualitave research) and
empirical studies (primarily quantave research).
2.2. Performing the Review. Following Wu et al. [18], this
study was conducted in two steps: idenficaon and coding.
In the first step, an arcle was selected to the potenal pool
when it qualified for either of two criteria: (a) the research
involved a specific AI technique as an intervenon in
assisng learning or teaching and (b) it provided empirical
evidence or in-depth analysis. As already noted, only arcles
indexed in SSCI were considered. It should be noted that
studies that focused on the development processes of AI
without educaonal implicaons or only adopted AI as a
learning subject without the employment of AI were
lOMoARcPSD| 59062190
Complexity 3
excluded from this review. Second, as for the analyc
studies, only studies that discussed the effect of AI
techniques on educaon were included. Each full text of all
the idenfied papers was read and screened individually by
three-panel members with doctoral degrees or
professorships in the field of learning technology. Studies
that did not fit clearly with the criteria were brought up for
panel discussion. The screening process yielded 100 arcles
out of the original set of 121.
In the second step, all the authors discussed themac
analysis principles and established a coding scheme in terms
of how AI was used in educaon. Two main categories were
invesgated: research quesons and technology adopon.
Firstly, with regard to research quesons, previous research
has found three basic models of AI in knowledge processing:
knowledge representaon, knowledge obtaining, and
knowledge derivaon [3]. Building on that foundaon, the
Table 1: The arcles coded by research queson, technology adopon, learning subject, educaonal level, research approach, and effects.
ID Authors
Research
queson
Technology adopon Learning Subject Educaonal level Approach Effects
1 Chin et al. [38] FEE Teachable agent Science educaon
58 6
th
grade students
(study 1); teachers and
134
5
th
grade students (study
2)
EXP OUT
2 Ngai et al. [55] AFF
Wearable biofeedback
circuit
Circuitry 7 to 9 grade students QE, SUR OUT
3 Wegerif et al. [42] REA
Intelligent
matchingpaern
algorithms
Online dialogue
100 undergraduates and
12 postgraduate students DA OTH+
4
Thomas and
Young [57]
GAM Adapve modelling Educaonal game 16 college graduates EXP, SUR PER
5 Yang et al. [27] ADP
Higher-order item
response algorithm
Elementary mathemacs 158 six graders in Taiwan EXP PER+
6 Moon et al. [58] GAM
Experience point data
modelling
Digital game 40 plays EXP PER+
7
McLaren et al.
[54]
AFF
Intelligent tutoring
system
Chemistry 132 high school students QE OUT
8 Jones [43]
Invesgang decentralized theory of arficial intelligence
Exploring creave thinking
9
Vaam et al. [24]
REA
Visualizaon Science educaon 157 middle school student
QE
OTH+
10
Jonassen [45]
Introducing an ask system: interacve learning system
11
Magnisalis et al.
[21]
Review of adapve and intelligent systems for collaborave learning support: adapve and intelligent systems
12
Albin-Clark et al.
[56]
ROL
GAM
4 early childhood lectures
Graphic simulaon Construcon EXP, SUR PER+
and many students
13
Wong and Looi
[59]
Exploring swarm intelligence
14
Seni [60]
Invesgang the relaonship between neurosciences and organizaonal cognion
15 Lin et al. [51] AFF
EXP,
Facial recognion Digital art course 20 adults PER
SUR, INT
16
Heslep [61]
Introspecon to the misunderstandings of AI in educaon movated by AI enthusiasts
17
Nguyen and
Yang [28]
MAT
About 500 Vietnamese
news on many kinds of
Extracon algorithm Language learning DA 0
mobile phone from 2009
lOMoARcPSD| 59062190
4 Complexity
to 2010
18 Tierney [22] REC
Five interviews were
Natural language conducted generang
Language learning INT 0
process over seven hours of
recordings
19
Lawler and
Rushby [4]
Interview with Rover Lawler to give comments on the effect of computer technique on AI in educaon
20
Tufekçi and K¨
ose¨ [34]
FEE
Constraint-based
modelling
Programming 120 university students EXP, SUR OUT+
21 Zipitria et al. [19] ADP
Automac discourse
measure
Language learning
17 summaries wrien in
Basque language
EXP PER+
22 Chin et al. [39] REA Teachable agent
Kit-based science
curriculum
153 fourth grade students QE OTH
23
Mukherjee et al.
[30]
FEE
Text-to-diagram
conversion
Reading
12 pupils; 4 teachers; 2
technical professionals; 2
nontechnical persons
QE OTH+
24 Jain et al. [41] FEE Visualizaon History
Two undergraduate
classes in computer
science
QE PER+
25
Higgins and
Heilman [62]
REC
Automated scoring
system
Language learning
Game team (not
menoned the
educaonal level)
0
26 Melo et al. [9] REC AFF
Computaonal
organizaon
Muldisciplines
148 students involved
were either in high school
or in early college years
EXP PER
27
Flogie and
Aberˇsek [13]
AFF
Transdisciplinary
pedagogy
Natural science
100 students in 7
th
, 8
th
,
9th grades
SUR OTH+
lOMoARcPSD| 59062190
Complexity 5
Table 1: Connued.
ID Authors
Research
Technology adopon Learning Subject
queson
Educaonal level Approach Effects
28
Rapanta and
Walton [40]
REA Argument map Emira and Spanish classes 205 university students EXP OTH+
29
Nabiyev et al.
[29]
Visualizaon
CLA Intelligent tutoring Mathemacal
system
476 moon problems
from 9
th
grade
mathemacs textbooks of DA 0
Turkish Ministry of Educaon
30
Loeckx [8]
Analyc essay of opportunies for AI used in educaonal data mining, adapve learning, and creavity
31
Horakov´ a et al.´
[3]
CLA
Comparison of
arficial networks,
classificaon,
regression trees, and
decision trees
Arficial neural networks 120 text fragments QE, DA 0
32 Ijaz et al. [14] IMM Virtual reality History
60 undergraduate
university students
QE PER+
33 Liu et al. [31] REC
Intelligent tutoring
system
Language learning
30 sports arcles
including 100 sentences
DA 0
34
Malik and
Ahmad [32]
DEE E-assessment system Engineering
243 student of 8th graders
EXP 0
35 Malik et al. [23] DEE
Query trend
assessment system
Language learning
16 quesons from
Microso Students’ QA
Corpus
DA 0
36 Peng [63]
CLA
REC
K-means algorithm,
PageRank algorithm
Online learning More than 700 scholars EXP 0
37
MacIntyre et al.
[64]
MAT Text minding soware Language learning
10 accomplished adult
musicians and dancers
INT 0
38
Aoun [65]
Book review in terms of importance and limitaon of AI in educaon
39
Williamson et al.
[33]
Discussing the importance of neuroscience in educaon
40
Munawar et al.
[35] FEE
Intelligent virtual
E-
laboratory environment 161 university students
laboratory
SUR OTH
41
Samarakou et al.
[47]
ADP
Learning system on
Telecommunicaon 28 students studying diagnosis,
assistance,
networks informacs
and evaluaon
EXP OTH+
42
Fenwick [12]
Pondering the transformaon of teacher’ professional roles
43
Kessler [44]
Analyc essay of AI in the language teaching
44 Pet et al. [36] GAM
Online educaonal
programming plaorm
Programming
400 students and 12
teachers
EXP OUT+
lOMoARcPSD| 59062190
6 Complexity
Table 1: Connued.
45 Kelly et al. [37] DEE
Queson authencity
measuring system
English and language arts
8
-
9 grade A large archive
database of text
transcripts of 451
observaons from 112
classrooms and 132 high-
quality audio recording
from 27 classroom
DA 0
46 Ge et al. [20]
CLA
MAT
Autonomous learning
system
Sports
Students of 2016 from a
college are selected for PE
tesng. Samples of the
150 quesons are
collected
DA 0
47 Sun [66] FEE Learning system Language learning
176 valid enterprise
quesonnaires and 178
student quesonnaires are
obtained
QE, SUR OTH
48
Auerbach et al.
[67]
IMM
Roboc hardware and
soware plaorm
Arficial Evoluon 42 postgraduate students EXP OUT
49
Boulet and
Durning [68]
Exploring online assessments system applied for the measurement in medical educaon
ID Authors
Research
Technology adopon
queson
Learning Subject Educaonal level Approach Effects
50
Cukurova et al.
[69]
Arficial intelligence
REA and mulmodal data
127 quesonnaires and 47 audio
recordings from
Debang skills SUR, DA OTH+
candidates who have applied
to become a tutor
51 Du Boulay [70]
Intelligent tutoring
ADP
systems (ITSs)
STEAM, conceptual
(Not menoned the understanding, and
EXP OTH
educaonal level) dialogue-based
learning
52
Hughes [71]
Review on papers about early self- and coregulaon from arficial intelligence perspecve included
53
Kay and
Kummerfeld [72]
Lifelong and life-wide (Not menoned the
FEE Learning system 0
personal user models educaonal level)
54
Kio and Knight
[73]
Invesgang three tensions in ethics when applying arficial intelligence and data analysis (AIDA) in educaon
55
Luckin and
Cukurova [74]
REA
Learning sciences- Problem-solving, learning driven
AI data, and debang
Data in one case from high
school
EXP,INT OUT
56
Sellar and Gulson
[75]
DEE AI and data science Educaon policy
4 semistructured
interviews with five
senior policymakers,
technical staff, and data
sciensts
INT 0
57 Sharma et al. [76]
ADP
Online adapve selfassessment procedure
Web technologies with mulmodal
data
Thirty-two undergraduate
students
EXP 0
58
Wang and Wang
[77]
Developing an arficial intelligence anxiety (AIA) scale
Exploring the relaonships between AIA and movated learning behaviour
lOMoARcPSD| 59062190
Complexity 7
Table 1: Connued.
59
Webb et al. [78]
Discussing how me and temporality are used and inflected with the introducon of AI in educaon policy
contexts
60
Williams [79]
Analyzing implicaons of arficial intelligence, data analycs, and blockchain technology for the academy
61 Williamson [80]
Learning analycs, AI,
FEE and other soware for Higher educaon Higher educaon SUR OUT+
data collecon
62
Winters et al. [81]
Invesgang the exisng digital structural violence and the approaches to tackling it
63 Rowe [82]
Exploring the effect on educaon reform brought by intangible economy which is shaped by globalized datasets
such as OECD PISA and arficial intelligence
64
Ally [83]
Idenfying the shaping forces for future educaon and competencies required by future digital teachers
65
Song and Wang
[84]
Analyzing analysis of worldwide educaonal arficial intelligence research development in recent twenty years
66 Ulum [85]
30 students from a state
FEE Versant English test English language learning DA,INT PER+
university in Turkey
67
Costa-Mendes
et al. [86]
Educaonal data Mullinear
regression
REA High school grades collecon from preschool, EXP OUT+
model primary, and high school
68 Zhai et al. [87]
Invesgang the factors impacng machine-human score agreements in machine learning-based science
assessments
69
Lous and
Madden [88]
First year students from
FEE Bayesian networks Internet of Things Bachelor of Science in EXP OUT+
Compung program
70
Breines and
Gallagher [89]
Introducing the applicaon cases of teacherbot in the University of Edinburgh
71 Campo et al. [90]
Middle, a Moodle plug-
REA in using a Bayesian Computer science 45 university students EXP OUT
network model
72
Papadopoulos
et al. [91]
Crically reviewing the research on the use of socially assisve robots (SARs) in the preterary classroom and its
benefits and disadvantages
73
Berendt et al. [92]
Examining benefits and risks of arficial intelligence (AI) in educaon in relaon to fundamental human rights
74
Standen et al.
[93]
Teachers selected learning material
from a library to
67 parcipants aged
ADP The MaTHiSiS system create their own learning EXP OTH
between 6 and 18 years acvies
and learning
graphs
ID Authors
Research
Technology adopon Learning Subject Educaonal level Approach Effects
queson
75 Liu et al. [94]
870 observaons have
been collected from 5
FEE BP neural network Undergraduate educaon EXP OUT+
consecuve academic
years in one university
76 Knox [6]
Analyzing the polical economy of arficial intelligence (AI) and educaon in China, with government policy and
private sector enterprise introduced
77 Cope et al. [95]
Students studying at the CGScholar
(Common
FEE Disciplinary knowledge masters and doctoral EXP OUT
Ground Scholar)
levels
lOMoARcPSD| 59062190
8 Complexity
Table 1: Connued.
78
Westera et al.
[96]
Reviewing the arficial intelligence (AI) for serious games, presenng reusable game AI comp
relevance for learning and teaching, AI approach, and applicaon cases
onents an
d their
79
Bonneton-Boe´
et al. [97] FEE
The Kaligo, a digital
Handwring Kindergarten
notebook applicaon
EXP OUT
80
Smutny and
Schreiberova [98]
ROL
(Not menoned the
Chatbots Disciplinary knowledge
educaonal level)
DA 0
81 Lucy et al. [99] DEE
Natural language (Not menoned the
History
processing educaonal level)
EXP OTH
82
Yakubu et al.
[100]
DEE
Arficial neural Learning management 1116 students in four network
(ANN) systems (LMS) Nigerian universies
SUR
PER
83
Bonami et al.
[101]
Analyzing th
e educaon through 21
st
-century skills and the impact of AI development in the age of plaorms,
taking research, applicaon, and evaluaon into consideraon
84
Koc-Januchta´
et al. [102]
DEE
Inquire Biology
24 students from
(arficial intelligence- Biology EXP,SUR OUT
Stockholm University enriched
textbook)
85
Tran and
Meacheam [118]
Introducing
four innovave projects that aim to extend learning management systems and improve the level of
automaon
86
Nye et al. [103]
DEE
MentorPal STEM 31 high school students SUR PER
87
Webb et al. [104]
Invesga
ng the implicaons of recent developments in machine learning for human learners and learning
88 Tsai et al. [105] DEE
3552 students from a
Deep neural networks
Disciplinary knowledge SUR 0
university in Taiwan
89
Alyahyan and
Dustegor [106]
Construcng guidelines to apply data mining techniques to predict student success
90
Renz and Hilbig
[107]
Analyzing th
e drivers and barriers that currently affect data
-
based teaching and learning paths from the
perspecve of EdTech companies
91
Gulson and
Witzenberger
[108]
Invesgang
how automated educaon governance assemblage includes new forms of experse and authority
and constutes EduTech as an important policy space
92
Kerimbayev et al.
[109]
Review the r
esearch aimed at studying robot
-man interacon, taking Russia and Kazakhstan as an example of the
internaonal cooperaon in the sphere of robocs
93 Fu et al. [110] FEE
AI-enabled learning
Language learning 15 language learners INT,SUR PER+ tools
94 Salas-Pilco [111]
Examining t
he use of arficial intelligence (AI) and robocs in lear
sciences
ning designs from the perspecve of learning
95 Yıldız [112] FEE
Conceptualizaon
SCM-AI performances
53 five
-
year
-
old and 49 seven
-
year
-
old Turkish
EXP OUT
monolingual children
from a primary school
96
Tolsgaard et al.
[113]
Crically rev
iewing the published ap
plicaon and potenal role of data science and machine learning in Health
Professions Educaon
97
Hsu [114]
DEE
AI Chatbot
English language learning 30 university students EXP OTH
98
Wu et al. [115]
CLA
Machine learning
classificaon model
A hybrid advanced stascs
24 university students EXP 0 course
99
Wang et al. [116]
ADP
Squirrel AI learning
Math 200 eighth grade students EXP OUT
lOMoARcPSD| 59062190
Complexity 9
Table 1: Connued.
100
Rybinski and
Kopciuszewska
[117]
AFF
Natural language
processing (NLP)
models
640,349 reviews of 132
Higher educaon EXP 0
universies
CLA: classificaon; MAT: matching; REC: recommendaon; DEE: deep learning; FEE: feedback; REA: reasoning; ADP: adapve learning; AFF: affecon
compung; ROL: role-playing; IMM: immersive learning; GAM: gamificaon; EXP: experiment; QE: quasiexperiment; DA: discourse analysis; INT: interview;
SUR: survey; OUT: outcome; PER: percepon; OTH: others including affecon, crical thinking, and creavity. : stascally significant change. +:
recognizable change without conducng significance tests. 0: focus on algorithms test without examinaon of learning performance.
lOMoARcPSD| 59062190
10 Complexity
research quesons of the sample papers were classified into
three dimensions: (a) development, focusing on the
knowledge presentaon model; (b) extracon, centering on
how to obtain knowledge from data mining; and (c)
applicaon, emphasizing the human-computer interacon
through informaon derivaon. Secondly, with regard to
technology adopon, the focus was on the types of
technology that the study adopted, which were further
categorized into soware (e.g., algorithms and programs)
and hardware (e.g., sensors and devices such as virtual
reality). It should be noted that a study with technology
without an AI purpose in educaon was not included. A
detailed descripon is shown in Table 1 and it includes
learning subject, educaonal level, research approach, and
effects. Moreover, the researchers conducted further
frequency comparisons on the associaons between the
research purposes and some factors such as AI technology
adopon as well as me periods to predict the trends and
challenges of AI in educaon.
3. Findings and Discussion
According to the above coding criteria and content analysis,
the three dimensions of research quesons are shown in
Table 2 and the 72 studies from 63 empirical studies (5
papers have two studies and 2 papers have three studies)
are further subclassified into 11 categories. There are 23
studies in the dimension of development. The AI technique
was ulized as a development tool for the construcon of a
smart learning environment, which can be subclassified as
focusing on the development of algorithms including
classificaon, matching, recommendaon, and deep
learning for teaching and learning purposes. Addionally, 35
reviewed studies were found in the dimension of extracon,
which referred to the applicaon of developed AI
techniques, normally based on algorithms, to offer students
feedback, reasoning, and adapve learning. 14 empirical
studies were found in the dimension of applicaon which
consisted of affecon compung, role-playing, immersive
learning, and gamificaon. In the integraon dimension, AI
techniques included those involving human factors as vital
variables to idenfy and analyze learners’ personalized
features. In such studies, human-computer interacon was
generated to improve such characteriscs as creavity,
responsibility, and crical thinking that can impact learners’
performances and percepons. The following secons
describe what educaonal issues were dealt with in the age
of AI and how AI technique was employed in each research
queson.
3.1. Dimension of Development. As shown in Table 2, 16
empirical studies were found focusing on the development
of educaon systems such as intelligent tutoring system
(ITS) and electronic assessment. The development
procedure was usually conducted with an inducon-
deducon approach, in which prior experiments and data
were analyzed to predict the variables followed by the
algorithm tesng to obtain the final modelling equaon
[19].
Generally, the development of an educaonal system is
constuted of three components: the presentaons, logical
modelling, and data dimension [20]. All the 23 studies
centered on logical modelling, while no study was found on
the presentaon methods or data mining. The possible
explanaon may due to that the modelling techniques were
the foundaon of AI technique and fundamentally
penetrate throughout the procedure of system
development. In this dimension, the research was generally
conducted in the domain of computer science or
informaon science, and the domain knowledge as the
source material was imported into algorithm frame (shown
in Figure 1(a)) with few pedagogical designs reported. For
example, Horakova et al. [3] aimed to explore the
classificaon ability of a text mining machine using three
classificaon techniques. The results show that arficial
neural networks (ANNs) were significantly more effecve
than regression trees and decision trees to separate
educaonal texts or text fragments.
Addionally, in terms of the matching/group formaon
modelling, prior research employing stereotype theory has
assessed that the Bayesian networks, associaon rules,
clustering, fuzzy C-means, and the fuzzy and genec
algorithms were well-accepted algorithms for the modelling
of individual properes of the student. These techniques
provide potenal indicaons for the invesgaon of forming
homogeneous and heterogeneous groups in an educaonal
context [21].
Moreover, the trends of the growing amount of data
challenge educators to analyze qualitave data efficiently.
Natural language processing (NLP) provided a means to
diagnose the problem and make a recommendaon by
simplifying and accelerang the discovery of what lies
within the data [22]. However, the assessment of a complex
educaonal system requires more profound informaon
retrieval. The integraon of mulple approaches, such as
benchmark in NLP/Semanc Web field, was suggested to
model smarter computer-aided systems in which agents
could be trained automacally [23].
To opmize the modelling in the learning context, the
hierarchical structures were considered as potenal
soluons to model the educaonal system. This is because
educaon is generally a complex system with the exhibion
of subsystems and components, in which the invisible causal
processes among subsystem/component behaviours would
causally affect each other [24]. It was suggested that
systemac modelling should analyze three dimensions in
the educaon context: learners variaon, learning
lOMoARcPSD| 59062190
Complexity 11
domains, and learning acvies [25, 26]. For example, some
researchers constructed the higher-order item response
theory framework involving the overall ability at the first
dimension and mulple domain abilies at the second
dimension, which has been well adopted in the automac
problem-solving process [27].
Based on the above and Nguyen and Yangs suggeson
[28], the aims of developing an AI-integrated system in
educaon could be grouped into four types: classificaon (5
studies), matching (3 studies), recommendaon (5 studies),
and deep learning (10 studies). (1) Classificaon refers to the
reconstrucon of knowledge bases, in which the materials
Table 2: The number of studies concerning AI in educaon from 2010 to 2020.
Quantave research topics
Development (N 23)
Classificaon
Matching
Recommendaon
1
1
2
1
2
1
2
1
1
1
5
3
5
Deep learning
2
1
1
6
10
Extracon (N 24)
Feedback
Reasoning
1
1 1
1
1
2
1
2
3
4
7
2
16
10
Adapve learning
1
1
1
4
2
9
Applicaon (N 12)
Affecon compung
Role-playing
Immersive learning
1
1
1
1
1
1
1
1
1
1
6
2
2
Gamificaon
1
2
1
4
Quantave research
4
6
3
3
5
1
2
8
7
13
20
72
Qualitave research
0
3
3
1
0
0
1
1
3
9
16
37
Total
4
9
6
4
5
1
3
9
10
22
36
109
(a) (b)
matching
Adapve learning
Reasoning
lOMoARcPSD| 59062190
12 Complexity
(c)
Figure 1: The hierarchy of arficial intelligence in educaonal implementaon. (a) The dimension of system development, (b) the
dimension of extracon, and (c) the dimension of applicaon.
could be categorized according to varied characteriscs.
Classificaon demarcates knowledge content, which
contributes to the accuracy of text analysis [3]. For example,
some researchers developed an ITS with the characteriscs
of categorizing moon problems, by which learners could
easily access different types of moon problems in
Mathemacs [29]. (2) Matching refers to a conversion
mechanism, in which varied sets of classificaon are
connected to specific learning purpose. For example, a text-
to-diagram system was developed for blind students to link
geometry words to an underlying diagram on the Braille
printout, which has been cerfied as an effecve
teaching/learning tool at a Blind school [30]. (3) The
recommendaon is regarded as an intelligent authoring
tool. With the support of the natural language process, it
could automacally create new themes, theories, and
pedagogical contents as a response to learners’ feedback, to
help teachers save me and effort [31]. It constructed a
human-computer interacon and widely used to generate
real-me and intelligent feedback according to learners’
input, which has been regarded as a reliable feature in
modern assessment system [32]. (4) Deep learning, or
machine learning, is a comprehensive approach of big data
processing and learning behaviour analysis. Based on the
proliferaon of big data in educaon, such as learning or
teaching behaviour, the system could self-adjust to meet
users’ dynamic requirements by upgrading its algorithms
[33].
To date, some studies have reported the lack of
significant impact on improving teaching. The challenge was
largely aributed to the weak pedagogical design and lack
of appropriate assessment criteria [8]. Future research
should therefore be grounded in learning theories so that
more acceptable, accessible, and efficacious AI can be an
integral part of learners’ lives.
3.2. Dimension of Extracon. Educators have begun to
explore suitable applicaons of AI techniques in their
teaching. There are currently some AI applicaons that have
achieved the integraon of technique, domain knowledge,
and pedagogical design. The three types of pedagogical
applicaons of AI idenfied in this review were feedback (16
studies), reasoning (10 studies), and adapve learning (9
studies). While these applicaons could be interlinked, they
were categorized as such based on the classificaon
explicated by the authors of the reviewed arcles.
3.2.1. Feedback. One of the challenges impairing
personalized learning is the inappropriate sequencing of
contents. The restructuring of presentaon sequences is
seeking a way to redefine the organizaon of knowledge
according to the student’s reacon. In this situaon,
feedback is an important approach to meet learners’
proximal learning paerns [9]. Using an arficial neural
network, the system provides immediate feedback
according to students’ input to help them gradually get
access to the abstract concepts and perform praccal
exercises. Besides, researchers perceived a posive trend
towards the system, which may aribute to two
perspecves.
(1) Based on Ohlsson’s theory, students can learn from
the feedback generated as the result of an error [34]. In a
physical teaching environment, the teacher could interact
with students immediately as difficules arise. It is,
however, difficult for such just-in-me interacon in an
online context. The situaon requires intelligent algorithms
to provide feedback automacally. For example, with the
help of pedagogical agent-based cognive architecture, the
intelligent virtual laboratory was developed to give
appropriate feedback to students who encounter difficules
in the laboratory [35]. Besides, a learning website, Jutge.org,
was developed with the features of a rich and well-
organized problem repository. The website provides instant
feedback and helps students to progressively solve
problems and learn from their mistakes [36]. (2) Immediate
feedback promotes acve training in interacve learning
lOMoARcPSD| 59062190
Complexity 13
environments that would benefit learners comprehension
diagnosis [19]. The previous study combined speech
recognion, natural language processing, and machine
learning to measure the quality of classroom talk, in which
new forms of interacon were created to provoke thoughts
and further shape the effecve interacon of the learning
environment [37]. Another AI system used path traversal
algorithms to establish causal chains, by which students
were provided with elaborated feedback and hints rather
than the correct answers. The learning-by-teaching context
was constructed by learners’ self-organizaon of
interacons and their interpretaon of feedback [38].
Although a large number of benefits were reported with
respect to automated feedback of domain knowledge, no
research in this review had established the connecon to
pedagogical theories. Most of the authors in the
development dimensions were from the computer science
domain, which leads to their focus on the presentaon of
source data (domain knowledge) technically without much
pedagogical consideraon.
3.2.2. AI-Supported Reasoning. The recursive feedback may
have the potenal to foster learnersabilies to reason in
specific ways because the human-computer interacon is
able to engender among the students a sense of
responsibility toward improving the construcon of
knowledge repository [39]. The reconstrucon of the
knowledge repository was seen as a process of using
modelling to realize pedagogical design as shown in Figure
1(b). However, some researchers found that novices such as
students and preservice teachers showed minimal
understanding of the invisible causal behaviours in the
system compared to experts and experienced teachers [24].
Another research showed a similar conclusion: students
were able to learn the relevant facts and pairwise relaons,
while they may sll fail to reason with them very well [39].
One possible explanaon could be that reasoning is largely
invisible and it is difficult to induce the processes of
reasoning through the observaon of the behaviours. AI
techniques such as the visualizaon technique could be
applied to foster learners’ reasoning.
To help learners improve their reasoning, the graph
structure [29] and learners’ engagement [24] techniques
have been studied. For the graph structure, intelligent
systems could be developed to make thinking visible. In a
sense, the simulaon approach of the AI technique was
employed to mimic thoughts tracking the reasoning visually
in real me. For example, the argument-mapping tools were
designed to assist learners with visualizaon of the premises
and conclusions of arguments. The findings showed that a
sequence of connected arguments was chained together for
learners to make an ulmate conclusion [40]. Drawing from
the sociocultural theories of learning in designing AI to
support students’ reasoning, Vaam et al. [24] reported that
engaged learners could beer understand the mulple
levels of organizaon in complex systems. Therefore,
students’ engagement is an essenal aspect to be
considered for the design of a learning system that aims to
support reasoning.
The hierarchical reasoning generated by the intelligent
system had beneficial effects on students’ learning. Firstly, it
may help learners to opmize the elucidaon of the
relaonships between the subcomponents of a parcular
topic. In return, the intelligent reasoning system can be used
as a form of evaluaon to assess if the student has captured
enough concepts for the given topic [41]. Secondly, the
system could provide an argumentave interacon which
placed great significance in the construcon of collaborave
learning atmosphere. It is because, as a result of peers’
reasoning, learners tend to externalize their arguments and
improve their premises. Jain et al. [41] combined visualized
mapping tool with collaboraon scripts. The design
successfully helped learners to analyze and evaluate
opposing posions on contenous topics. Generally,
researchers regarded the reasoning visualizaon tools as
valuable scaffolds to develop learners’ crical thinking and
wring [40].
However, using AI techniques, including visualizaon
and hierarchical reasoning modelling, may be inadequate to
support reasoning. The four studies reviewed focused on
the ulizaon of modelling to support general reasoning,
while the reasoning model should be largely domain-
specific [24, 39, 40, 42]. Moreover, there is an unresolved
challenge in coding learners’ behaviours as far as AI-
supported reasoning is concerned. The reasoning process
may be more effecve when learners’ personalized
performance is considered. Although the visualized
reasoning tools could perform well in a small-scale group
seng, it is difficult to obtain adequate reasoning analysis
of the data from a large populaon because the reasoning
system fails to adjust itself automacally. Therefore, the
requirements of dealing with increasingly large and diverse
data demand self-adapve alternaves [9].
3.2.3. Adapve Learning. Based on the new decentralized
theories of AI and social cognion, the apparent complexity
of learners’ behaviour was largely a reflecon of the
complexity of the learning environments. This prompted
educators to provide adapve scaffolds for diversified
learning environments with various types of learners.
Different from the feedback system that oers stock
responses, the adapve educaonal system is a formave
and correcve automated system that can adjust itself
(target of intervenon) to suit individual learners’
lOMoARcPSD| 59062190
14 Complexity
characteriscs, needs, and preferences (pedagogical
objecve) [43]. Although only three empirical studies were
idenfied in this review, some researchers were very
posive to the future promoon of adapve system in
teaching and learning. Technologies such as intelligent
speech recognion and automated wring evaluaon [44]
have been tested with promising findings. In addion, there
was substanal evidence showing that adapve intelligence
enhances learning by automacally enabling learners to
locate and access proximal educaonal resources with
respect to navigaon and presentaon support [45].
Previous research has emphasized that the design
dimension was a worth exploring alternave in the
applicaon of adapve system [46]. To design successful
adapve systems in educaon, curriculum designers and
system designers have to leverage on to include the
modelling of the problem-solving process in the specific
domain knowledge and the use of big data [21, 44]. Firstly,
the mechanism of the adapve system connects learners’
prior domain knowledge and the evaluaon of their current
domain performance to scaffold their problem-solving [47].
In parcular, the pedagogical design is essenal in adapve
intelligent context. It involves the selecon of adapve
algorithms and consideraons about the compability of
the learning style and the intelligence supporve methods.
In this sense, the assumpon that AI would threaten the
teachers’ posion may be unfounded because of teachers’
vital role as curriculum designers. Secondly, the adapve
system is empowered by big data. Since the main feature of
the adapve learning system is personalizaon,
accumulaon of big data such as the range of diverse
individual characteriscs and learning style and preferences
is necessary for intelligent personalizaon to be realized.
However, research on personalizaon in the context of the
adapve system is limited to the users’ characteriscs
related to domain knowledge. The deeper internal
characters, such as human mental status and creavity,
were barely noced and studied [21]. This however has vital
research potenal with the development of advanced AI
techniques such as biofeedback techniques.
3.3. Dimension of Applicaon
3.3.1. Technology Adopon in the Applicaon Dimension.
The dimension of applicaon highlights the importance of
including human affecon in the applicaon of AI in
educaon. The latest research has indicated that affecon
had increasingly been reported to exert a significant
influence on decision-making, percepon, and learning
[48]. Previous studies on the measurement of learning
performance only focused on two dimensions: learning
outcomes (e.g., scoring and achievement) and percepons
(e.g., sasfacon and acceptance), whereas other aspects
were less noced. Based on the maturity of biofeedback
technique, such as eyetracking and EEG, affecon
compung was increasingly adopted to invesgate
students' internal movaons on learning, such as creavity
and responsibility [49, 50].
According to the content analysis of the selected papers
shown in Table 1, there are five typical AI techniques that
supported affecon compung and analysis in the
educaon sector. They are complex algorithms,
visualizaon, XR (virtual/augmented/mixed reality),
wearable technique, and neuroscience. In many situaons,
they supported each other to construct a smart learning
environment and system. (1) Complex algorithms were
designed with consideraon of human factors rather than
the simple combinaon of funconal blocks. From the
perspecve of human-computer interacon, the learners
should be treated as a knowledge creator rather than the
receiver, which helps to generate posive affecon status.
From the perspecve of presentaon modes, the tradional
declarave statements in a computer system should be
replaced by more diversified verbal presentaons such as
dialogue, coaching, and generality. (2) Visualizaon was
seen as an opmal method chosen for the soluon of
complex concepon. One of the benefits of visualizaon is
making complex knowledge entertaining, such as game-
based learning, in which learners’ movaon will be greatly
generated. (3) XR including virtual/augmented/mixed
reality provides a highly simulated learning context, which
may be challenging to realize in physical classrooms. For
example, to help learners understand complex landforms in
geography, XR indulges students into a lively and creave
status. (4) The wearable technique, such as Google glasses,
helps to integrate learning acvity into somatosensory
moves. Although it was sll in an exploratory period, it has
great potenal to advance domain knowledge in a praccal
context in daily life. (5) Modern neuroscience exploits how
the brain works and this expands the research of learning to
include the learners physiological state. Research in this
area would enrich understanding about individual variaons
and could provide addional avenues to match instrucon
with the most opmal guidance.
3.3.2. The Categories of the Applicaon Dimension. With the
supports of the above five AI techniques, four types of
learning models were generated with the applicaon of
affecon analysis, which was biofeedback (6 studies),
roleplaying (2 studies), immersive learning (2 studies), and
gamificaon (4 studies).
Affecon compung refers to the analysis of human
emoons and feelings captured by physical sensors and
affecve algorithms, which has gained much aenon in
recent years. Affecon compung enhanced human-
lOMoARcPSD| 59062190
Complexity 15
computer interacon. Based on the facial idenficaon,
some researchers improved the intelligent tutoring system
by which students’ emoonal status was detected to give
them mely emoonal feedback [51]. Two essenal aspects
are needed to opmize the affecon compung technique:
first, teachers have to make mely appropriate instruconal
adjustments according to learners’ affecve status; second,
comprehensive operaon of mulmode affecon sources as
a single source is unlikely to provide accurate analysis of
affecon. For example, the eye-tracking technique could
capture learners’ eye fixaon to track the aended area, but
the reasons for the foci may be aributed to different
affecons such as interest, anxiety, or even distracon. An
addional source of data such as EEG could help to make a
more accurate assessment [52].
Role-play is a learning method that inspires students to
ponder on problems with affecons assuming varied roles.
Some algorithms were designed with the integraon of
roleplay into the pedagogical design, where students are
taught by an intelligent agent rather than being taught by
the learning system [39]. Enlisng role-play can enhance
learners’ investment in their interacons with computers.
More than that, learners’ sense of responsibility was
exerted towards the intelligent agent, which was consistent
with the research from Chase et al., demonstrang that
students may work harder on behalf of their agents than
they would for themselves [53]. Addionally, to movate
students to act as a companion to an intelligent agent, the
politeness presentaon mode was employed in the
intelligent tutoring systems, which was observed to benefit
the needy students [54]. The future research of role-play
may focus on granng access to students so that they could
customize their roles and target agents.
Immersive learning is an approach that enables students
to customize scenes of characters engaging in full-view
learning sengs. The enhancement of XR, 3D graphics, and
wearable devices could promote the learning performance
and these are strongly related to immersive affecon, which
generated students’ academic performance and posive
percepons, such as excitement, enthusiasm, and creavity.
For example, learners could obtain a high degree of
excitement in the immersive learning environment.
Immersive environment can also be coupled with immersive
collaboraon with gestures, emoons, and nonverbal
communicaon [14]. Using immersive learning may also
reduce students’ sense of being inmidated by complex
topics and technical concepts when they expose to
simulated technological and compung issues [55]. Most
importantly, many immersive learning tools encourage
learners’ enthusiasm to create and change the
environments, which could foster creavity [56]. However,
few studies were found to consider domain knowledge as a
variable. The possible reason may be that many immersive
learning tools were in the explorave stage. Further
invesgaons in specific domains are eagerly needed.
Gamificaon has emerged as an important theorecal
noon in the educaon sector. The most successful
educaonal games ghtly integrate the pedagogical design,
domain knowledge, and affecon elements with gameplay.
AI has assisted the integraon of the game and knowledge
domain, and the further potenal is making the game adapt
to the learnersbehaviours and affecons dynamically [57].
One of the examples appropriately integrang domain
knowledge with affecon is Minecra Edu. This is a historical
simulaon game where students can learn about historical
figures and events or get insight into the spread of
epidemics. Learners could get access to historical events
with authenc emoons in the real-me interacon, and
the collateral emoon would help them beer understand
the specific content knowledge [8].
Another example employed a game reward system as
movaonal mechanisms to promote voluntary and
proacve learning. The results showed that the reward
system had a desirable fit with the pedagogical design, and
the future educaonal algorithms might beer get
associated with the field of arficial intelligence to movate
emergent learning [58].
3.4. The Results from Qualitave Research. According to
selected qualitave research (as shown in Table 3), the
exploraon of AI in educaon experienced a process from
theorecal research to a specific pracce field, and at last
back to review. Simultaneously, qualitave research also
provided support for the development of quantave
research throughout the whole process. Some theorecal
studies were at the forefront. For example, in 2011 and
2012, qualitave research on decentralized theory [43] and
swarm intelligence [59] appeared, and then the real
arficial intelligence research began. AI algorithms were not
very mature at the beginning while advanced intelligent
algorithms are usually based on big data technology, and
they could constantly learn and improve in the massive
data. The big data must be decentralized and group-
oriented. Therefore, we believe that the early theorecal
research has played a significant supporng role. In 2019,
researchers aached more emphasis on the summary of
previous studies and prospects for future development, and
more consideraon will be given to the status quo, future,
and possible problems of AI in various sectors of educaon.
4. The Research Trends of AI in Education
4.1. Technology Adopon of Internet of Things. The exisng
research mainly focused on the virtual online system, and
the Internet of Things (IoT) is less noced. Learners’
biofeedback also needs to be explored in future educaonal
lOMoARcPSD| 59062190
16 Complexity
research. According to the reviewed papers, a majority of AI
technology in educaon focused on online informaon
technology or system (107 out of 109), such as intelligent
tutoring system, intelligent virtual laboratory, and
assessment system. Only one study [55] employed a
wearable circuit to examine learners’ biofeedback. This may
be aributed to the fact that the intelligent online system is
well established, easier to build on, and cost-effecve.
However, to cater to diverse learning contents and varied
learning skills, the IoT holds much promise. It may enhance
students’ spaal and mechanical understanding of physical
construcon processes in science educaon. The IoT
technology can simulate brain funcons in physical context
to sense and understand human’s cognive behaviours,
which apparently opmizes human cognion and
performance in two qualitave studies [33, 60]. Although no
empirical studies in the selected papers were found to test
the effect of IoT technique on educaon, the IoT with
affordable costs and wearable compung devices could be
a potenal area of future development of AI in educaon.
This is consistent with the Horizon report in 2019.
Table 3: Qualitave research topics.
2020
(1) AI research development in recent twenty years
(2) Machine learning-based science assessments
(3) Applicaon cases of teacherbot
(4) Socially assisve robots (SARs)
(5) AIED in relaon to fundamental human rights
(6) Polical economy of AI and educaon
(7) AI for serious games
(8) Impact of AI development in the age of plaorms
(9) Innovave projects extending LMS
(10) Machine learning
(11) Data mining techniques
(12) Data-based teaching and learning paths
(13) Automated educaon governance assemblage
(14) Robot-human interacon
(15) AI and robocs in learning designs
(16) Data science and machine learning
2019
(1) Assessment system
(2) Early self- and coregulaon from AI perspecve
(3) Ethical tension about applying AIDA in educaon
(4) Arficial Intelligence Anxiety (AIA) Scale
(5) AI in educaon policy contexts
(6) AI, data analycs, and blockchain technology
(7) Digital structural violence
(8) Intangible economy
(9) Future educaon and digital teachers
2018
(2) Neurosciences in Edu
(3) Misunderstandings of AI in Edu
2011
(1) Decentralized theory
(2) Interacve learning
4.2.SwarmIntelligenceinEducaon. Swarm intelligence has
become a vital development direcon of AI, where the roles
of teachers and students will be disrupvely changed.
According to the selected papers, the decentralized theory
was firstly invesgated in educaon in 2011 [43], followed
by the introducon of swarm intelligence in educaon in
2012 [59]. However, no empirical study has explored how
teachers and students meet the challenges brought by
swarm intelligence. It is predicted that the following two
topics may become the research trends according to the
features of swarm intelligence. Firstly, swarm intelligence
does not rely on centralized control of individual
behaviours. In this situaon, learners change from
knowledge absorbers to creators. They acvely constructed
knowledge by interfacing with the system in a variety of
contexts. Teachers’ “authories” may be challenged by a
group of experienced praconers such as engineers and
farmers, and a collaborave curriculum design would be
constructed by swarm intelligence system [45]. Moreover,
swarm intelligence may change teachers’ dues from
knowledge transmission to knowledge organizaon.
Previous research has suggested the exploraon of
crowdfunding or crowdsourcing by teachers on educaon,
and how teachers perform their organizing ability in the
future [5]. However, as Figure 2 presents, the invesgaon
from teachers’ perspecve is sll inadequate, which needs
further study. Secondly, swarm intelligence facilitated
adapvity in dynamic or unstable environments. Swarm
agents usually exchange informaon by leaving marks and
observing the acvies of their peers. For example, the best
soluon in the current moment may become unavailable in
the next moment. Therefore, it is suggested to invest further
how AI performs dynamic recommendaon for students on
different learning progress [59].
Neuroscience in educaon
Educaonal data mining, adapve learning, and creavity
Swarm intelligence
lOMoARcPSD| 59062190
Complexity 17
4. 3.DeepLearningandNeurocomputaon. Deep
learning or machine learning will reshape the interacons
between human beings and machines in the future. The
trends of human-computer interacon will no longer be
based on the perspecve of machine operaon by a human.
Instead, the machine can improve predicons by learning
from big data without being specifically programmed. Two
studies on deep learning were first menoned in the
selected papers in 2017 [23, 32]. In 2018, one empirical
study [37] was published and it focused the deep learning
technology on the modelling of scoring-based data.
However, the data based on human’s physical features were
less noced. Based on the basis of neuroscienfic
understanding of the brain, Pearson and IBM have proposed
to invesgate neurocomputaon brain-based educaonal
technologies [33]. However, only two qualitave studies
[33, 60] suggested the integraon of neuroscience and AI in
the educaon sector. Future research trends in integrang
brain funcon with deep learning techniques to opmize
human-computer interacon could be expected. It will
influence the applicaon and integraon of AI in educaon,
such as adapve learning and role-play. This view has been
reported in the Horizon report in 2018. Specifically, the
report forecasts that adapve learning techniques will be
further generalized in two to three years.
4.4. Evaluaon of AI in Educaon. All empirical studies
reviewed presented the posive effects of AI techniques on
educaon (see Table 1). However, the interview and the
review paper have, respecvely, surfaced the challenges or
misunderstanding of AI in educaon [4, 21]. There is a need
to arculate a holisc evaluaon criterion to measure the
effecveness of AI in educaon. To ensure the validity and
reliability of the evaluaon, a muldimensional model
should be adopted, which includes technique, pedagogical
design, domain knowledge, and human factors. Woolfs
[119] Roadmap for Educaon Technology predicted that in
the era of AI Educaonal Data Mining, the lifelong
assessment of students’ knowledge, their progress, and the
environments where they learn, as well as the success and
failure in teaching strategies, can be chronologically tracked.
Besides, current research is disproporonately focused
on specific educaonal contexts and a handful of variables.
As shown in Figure 2, most research sampled students as
parcipants, while teachers and professor praconers
were less noced; addionally, most researchers considered
science, humanity, and social science as subjects, but less
aenon was paid to sports, arts, and special educaon. For
example, only one study was found to develop text-to-
diagram conversion as a novel teaching aid for blind learners
[30].
5. The Challenges AI Confronted in Education
AI is a promising eld that faces many technology
bolenecks. The challenges would be more complex and
intricate, especially when they are connected to an
applicaon in educaon. The challenges this review
idenfies could be classified into three categories:
technique, teachers and students, and social ethics.
Although AI techniques displayed and predicted smart
computaon in the educaon domain, they generally fail to
bring “added-value” to large-scale students because of the
concern of costs, and the mainstream is sll occupied by
“basic value” [38]. Specifically, some researchers found that
many AI techniques were designed for a general situaon
that could not address the needs of a parcular domain,
specific learning acvies, or teaching goals. This would
prevent the actualizaon of personalized learning
experiences [8, 120].
Another great challenge reported in the Horizon report
in 2018 is the reconceptualizaon of the role of educators.
Teachers’ atudes towards AI have a significant influence
on the effecveness of using AI in educaon. Teachers may
swing from total resistance to overreliance. The former
could arise from inadequate, inappropriate, irrelevant, or
outdated professional development. The laer may be due
to teachers’ unrealisc expectaons. These teachers may
focus too much on the emerging AI technologies rather than
learning itself [44]. Addionally, from the perspecve of
students, AI technique may provide smart and efficient tools
that cause students to avoid doing the knowledge
processing work that teachers expect them to do. For
example, the AI translators may offer ready-made
illustraons, pronunciaon, fixed phrases, and even a serial
of examples. Students are thus unwilling to engage in the
inquiry processes that facilitate deep learning.
The ethical issues brought by AI are also challenging for
both researchers and educaonal praconers. It was clear
that AI has made great strides over the past few years,
mostly because of cheaper processing and the availability of
data; however, individual student data may be exposed,
shared, or used inappropriately. It is a constantly mindful
challenge that educators and AI engineers will face when
considering how we access, evaluate, and share the big data
and the results of data analysis [44, 65]. Another ethical
debate was conspicuously found in gamificaon that
emphasis should be put on learning and tend to “suck the
fun outof games, or on gameplay suck out the learning
[57].
lOMoARcPSD| 59062190
18 Complexity
6. Conclusions
Given the rapid growth of AI, there is an urgent need to
understand how educators can best ulize AI techniques for
the academic success of students. This paper reviewed AI in
educaon research from 2010 to 2020. It is found that the
research to date could be classified into three dimensions:
the dimension of development including classificaon,
matching, recommendaon, and deep learning; the
dimension of an extracon involving feedback, reasoning,
and adapve learning; and the dimension of applicaon
including affecon compung, role-playing, immersive
learning, and gamificaon. Moreover, based on the research
quesons and the related AI techniques, four research
trends were idenfied. They are the Internet of Things,
swarm intelligence, deep learning, and neuroscience, as
well as an assessment of the effect of AI in educaon. The
challenges of AI in educaon were also conspicuously seen
in terms of technique perspecve, teachers’ and students’
roles, and social ethical issues. These findings could be
valuable references for educaonal researchers, students,
and AI developers who plan to contribute to the relevant
studies. Furthermore, it seems clear that educators need to
work with AI engineers to address the gaps between
technique and pedagogy.
7. Limitations and Future Study
Although this review does propose some valuable trends
and potenal research direcons for AI in educaon, there
exist several limitaons. Firstly, the papers reviewed in this
study were filtered from Social Science Citaon Index, while
other databases on natural science (e.g., SCOPUS and EI)
and sources (e.g., reports, news, conference papers, and
patents) could be involved to offer a more comprehensive
overview in this eld. For instance, arcles from the
Internaonal Journal of Arficial Intelligence in Educaon
that has published 30 volumes were not considered. This
review therefore is limited only to SSCI arcles. Addionally,
the inial search could be extended using more keywords
such as adapve learning and tutor system, which may lead
to the latest technical reports of AI in educaon that were
not included in this paper. Secondly, since the current
review was not aempted to be inclusive but to provide a
systemac overview of AI in educaon, the analysis in this
review may provide a framework for future research
integraon. For example, a more formal meta-analysis could
be conducted on selected empirical studies that reported
effect sizes to see what impact on learning AI might be
having. Besides, the future analysis could go back further in
me to see if there were changes about the me that AI 2.0
started to make headways into educaon.
Data Availability
The content analysis data used to support the ndings of
this study are included within the arcle.
Conflicts of Interest
The authors declare that they have no conflicts of interest
to report regarding the present study.
Acknowledgments
This research work was supported by the 2020 Humanies
and Social Science Projects of the Ministry of Educaon
(Grant ID: 20YJC880118), Naonal Science Funding of China
(Grant ID: 61977057), 2019 Naonal Social Science Funding
of China (19ZDA364), and the project of Informazaon
Capability in University Governance System, Chinese
Associaon of Higher Educaon, 2020 (Grant no.
2020ZDWT18).
students students social sciences
Educaonal level Subjects
(a) (b)
Figure 2: The number of reviewed studies by educaonal level and subjects.
Others
College
lOMoARcPSD| 59062190
Complexity 19
References
[1] K. Kumar and G. S. M. Thakur, Advanced applicaons of
neural networks and arficial intelligence: A review,
Internaonal journal of informaon technology and
computer science, vol. 4, no. 6, pp. 57–68, 2012.
[2] J. M. Spector and D. J. Muraida, Automang Instruconal
Design: Concepts and Issues, Educaonal Technology
Publicaons, Englewood Cliffs, NJ, USA, 1993.
[3] T. Horakova, M. Houska, and L. Domeova, “Classificaon of
the educaonal texts styles with the methods of arficial
intelligence,Journal of Balc Science Educaon, vol. 16,
no. 3, pp. 324–336, 2017.
[4] R. W. Lawler and N. Rushby, An interview with Robert
Lawler,Brish Journal of Educaonal Technology, vol. 44,
no. 1, pp. 20–30, 2013.
[5] Dai, C. S. Chai, P. Y. Lin et al., “Promong students’
wellbeing by developing their readiness for the arficial
intelligence age,Sustainability, vol. 12, no. 16, pp. 1–15,
2020.
[6] J. Knox, Arficial intelligence and educaon in China,
Learning, Media and Technology, vol. 45, no. 3, pp. 1–14,
2020.
[7] A. Seldon and O. Abidoye, The Fourth Educaon Revoluon,
pp. 1–14, University of Buckingham Press, London, UK,
2018.
[8] J. Loeckx, “Blurring boundaries in educaon: context and
impact of MOOCs,The Internaonal Review of Research in
Open and Distributed Learning, vol. 17, no. 3, pp. 92–121,
2016.
[9] F. R. Melo, E. L. Flores, S. D. Carvalho, R. A. G. Teixeira,ˆ L. F.
B. Loja, and R. de Sousa Gomide, “Computaonal
organizaon of didacc contents for personalized virtual
learning environments, Computers & Educaon, vol. 79,
pp. 126–137, 2014.
[10] B. Boulay, Arficial intelligence as an effecve classroom
assistant,IEEE Intelligent Systems, vol. 31, no. 6, pp. 76–
81, 2016.
[11] M. Lacity and L. P. Willcocks, Roboc Process Automaon
and Risk Migaon: The Definive Guide, SB Publishing,
Ashford, UK, 2017.
[12] T. Fenwick, “Pondering purposes, propelling forwards,
Studies in Connuing Educaon, vol. 40, no. 3, pp. 367–380,
2018.
[13] A. Flogie and B. Aberˇsek, Transdisciplinary approach of
science, technology, engineering and mathemacs
educaon,Journal of Balc Science Educaon, vol. 14, no.
6, pp. 779–790, 2015.
[14] K. Ijaz, A. Bogdanovych, and T. Trescak, “Virtual worlds vs
books and videos in history educaon, Interacve
Learning Environments, vol. 25, no. 7, pp. 904–929, 2017.
[15] B. Kitchenham, P. Pretorius, D. Budgen et al., “Systemac
literature reviews in soware engineering–a terary study,
Informaon and soware technology, vol. 52, no. 8, pp.
792–805, 2010.
[16] R. Trescak, B. Yang, E. Zio, and X. Chen, Arficial
intelligence for fault diagnosis of rotang machinery: a
review, Mechanical Systems and Signal Processing, vol.
108, pp. 33– 47, 2018.
[17] Z. Wang and R. S. Srinivasan, A review of arficial
intelligence based building energy use predicon:
contrasng the capabilies of single and ensemble
predicon models, Renewable and Sustainable Energy
Reviews, vol. 75, pp. 796–808, 2017.
[18] Y.-T. Wu, H.-T. Hou, F.-K. Hwang et al., A review of
intervenon studies on technology-assisted instrucon
from 2005–2010, Journal of Educaonal Technology &
Society, vol. 16, no. 3, pp. 191–203, 2013.
[19] I. Zipitria, A. Arruarte, and J. Elorriaga, “Discourse measures
for Basque summary grading, Interacve Learning
Environments, vol. 21, no. 6, pp. 528–547, 2013.
[20] X. Ge, Y. Yin, and S. Feng, Applicaon research of computer
arficial intelligence in college student sports autonomous
learning,Kuram Ve Uygulamada Egim Bilimleri, vol. 18,
no. 5, pp. 2143–2154, 2018.
[21] I. Magnisalis, S. Demetriadis, and A. Karakostas, Adapve
and intelligent systems for collaborave learning support: a
review of the field, IEEE Transacons on Learning
Technologies, vol. 4, no. 1, pp. 5–20, 2011.
[22] P. Tierney, A qualitave analysis framework using natural
language processing and graph theory,The Internaonal
Review of Research in Open and Distributed Learning, vol.
13, no. 5, pp. 173–189, 2012.
[23] K. R. Malik, R. R. Mir, M. Farhan, T. Rafiq, and M. Aslam,
“Student query trend assessment with semancal
annotaon and arficial intelligent mul-agents,Eurasia
Journal of Mathemacs, Science and Technology Educaon,
vol. 13, no. 7, pp. 3893–3917, 2017.
[24] S. Vaam, A. K. Goel, S. Rugaber et al., “Understanding
complex natural systems by arculang structure-
behaviorfuncon models, Educaonal Technology &
Society, vol. 14, no. 1, pp. 66–81, 2011.
[25] A. Casamayor, A. Amandi, and M. Campo, “Intelligent
assistance for teachers in collaborave E-learning
environments,Computers & Educaon, vol. 53, no. 4, pp.
1147–1154, 2009.
[26] A. Gogoulou, E. Gouli, and M. Grigoriadou, Adapng and
personalizing the communicaon in a synchronous
communicaon tool, Journal of Computer Assisted
Learning, vol. 24, no. 3, pp. 203–216, 2008.
[27] C.-W. Yang, B.-C. Kuo, and C.-H. Liao, A HO-IRT based
diagnosc assessment system with constructed response
items, Turkish Online Journal of Educaonal
TechnologyTOJET, vol. 10, no. 4, pp. 46–51, 2011.
[28] B.-A. Nguyen and D.-L. Yang, A semi-automac approach
to construct Vietnamese ontology from online text,The
Internaonal Review of Research in Open and Distributed
Learning, vol. 13, no. 5, pp. 148–172, 2012.
[29] V. V. Nabiyev, U. Çakirog˘lu, H. Karal, A. K. Eru¨mit, and¨ A.
Çebi, “Applicaon of graph theory in an intelligent tutoring
system for solving mathemacal word problems,Eurasia
Journal of Mathemacs, Science & Technology Educaon,
vol. 12, no. 4, pp. 687–701, 2016.
lOMoARcPSD| 59062190
20 Complexity
[30] A. Mukherjee, U. Garain, and A. Biswas, “Experimenng
with automac text-to-diagram conversion: a novel
teaching aid for the blind people,Journal of Educaonal
Technology & Society, vol. 17, no. 3, pp. 40–53, 2014.
[31] M. Liu, V. Rus, and L. Liu, Automac Chinese factual
queson generaon, IEEE Transacons on Learning
Technologies, vol. 10, no. 2, pp. 194–204, 2017.
[32] K. R. Malik and T. Ahmad, “E-assessment data compability
resoluon methodology with bidireconal data
transformaon,EURASIA Journal of Mathemacs, Science
& Technology Educaon, vol. 13, no. 7, pp. 3969–3991,
2017.
[33] B. Williamson, J. Pyke, and S. Nemorin, “Biosocial spaces
and neurocomputaonal governance: brain-based and
brain-targeted technologies in educaon, Discourse:
Studies in the Cultural Polics of Educaon, vol. 39, no. 2,
pp. 258–275, 2018.
[34] A. Tu¨fekçi and U. Ko¨se, Development of an arficial
intelligence based soware system on teaching computer
programming and evaluaon of the system, Haceepe
Universitesi¨ Eg˘im Faku¨ltesi Dergisi, vol. 28, no. 2, pp.
469–481, 2013.
[35] S. Munawar, S. K. Toor, M. Aslam, and M. Hamid, “Move to
smart learning environment: exploratory research of
challenges in computer laboratory and design intelligent
virtual laboratory for eLearning technology, Eurasia
Journal of Mathemacs, Science and Technology Educaon,
vol. 14, no. 5, pp. 1645–1662, 2018.
[36] J. Pet, S. Roura, J. Carmona et al., Jutge.org:
characteriscs and experiences, IEEE Transacons on
Learning Technologies, vol. 11, no. 3, pp. 321–333, 2018.
[37] S. Kelly, A. M. Olney, P. Donnelly, M. Nystrand, and S. K.
D’Mello, Automacally measuring queson authencity in
real-world classrooms, Educaonal Researcher, vol. 47,
no. 7, pp. 451–464, 2018.
[38] D. B. Chin, I. M. Dohmen, B. H. Cheng, M. A. Oppezzo, C. C.
Chase, and D. L. Schwartz, “Preparing students for future
learning with teachable agents, Educaonal Technology
Research and Development, vol. 58, no. 6, pp. 649–669,
2010.
[39] D. B. Chin, I. M. Dohmen, and D. L. Schwartz, “Young
children can learn scienfic reasoning with teachable
agents,IEEE Transacons on Learning Technologies, vol. 6,
no. 3, pp. 248–257, 2013.
[40] C. Rapanta and D. Walton, The use of argument maps as
an assessment tool in higher educaon, Internaonal
Journal of Educaonal Research, vol. 79, pp. 211–221,
2016.
[41] G. P. Jain, V. P. Gurupur, J.
L. Schroeder, and
E. D. Faulkenberry, Arficial intelligence-based student
learning evaluaon: a concept map-based approach for
analyzing a students understanding of a topic, IEEE
Transacons on Learning Technologies, vol. 7, no. 3, pp.
267–279, 2014.
[42] R. Wegerif, B. M. McLaren, M. Chamrada et al., “Exploring
creave thinking in graphically mediated synchronous
dialogues,Computers & Educaon, vol. 54, no. 3, pp. 613–
621, 2010.
[43] A. Jones, “Philosophical and socio-cognive foundaons for
teaching in higher educaon through collaborave
approaches to student learning, Educaonal Philosophy
and Theory, vol. 43, no. 9, pp. 997–1011, 2011.
[44] G. Kessler, Technology and the future of language
teaching,Foreign Language Annals, vol. 51, no. 1, pp. 205–
218, 2018.
[45] D. H. Jonassen, Ask systems: interrogave access to
mulple ways of thinking, Educaonal Technology
Research and Development, vol. 59, no. 1, pp. 159–175,
2011.
[46] E. Walker, N. Rummel, and K. R. Koedinger, “Beyond explicit
feedback: new direcons in adapve collaborave learning
support, in Proceedings of the 9th Internaonal
Conference on Computer Supported Collaborave Learning,
vol. 1, Maratea, Italy, June 2009.
[47] M. Samarakou, G. Tsaganou, and A. Papadakis, An E-
learning system for extracng text comprehension and
learning style characteriscs, Journal of Educaonal
Technology & Society, vol. 21, no. 1, pp. 126–136, 2018.
[48] M. B. Ammar, M. Neji, A. M. Alimi, and G. Gouarderes,
The` affecve tutoring system, Expert Systems with
Applicaons, vol. 37, no. 4, pp. 3013–3023, 2010.
[49] I. Arroyo, D. G. Cooper, W. Burleson, B. P. Woolf, K. Muldner,
and R. Christopherson, “Emoon sensors go to school,
Arficial Intelligence in Educaon, vol. 200, pp. 17–24,
2009.
[50] K. Floyd, J. A. Hess, L. A. Miczo, K. K. Halone, A. C.
Mikkelson, and K. J. Tusing, “Human affecon exchange:
VIII. Further evidence of the benefits of expressed
affecon, Communicaon Quarterly, vol. 53, no. 3, pp.
285–303, 2005.
[51] H.-C. K. Lin, C.-H. Wang, C.-J. Chao, and M.-K. Chien,
“Employing textual and facial emoon recognion to
design
an affecve tutoring system, Turkish Online Journal of
Educaonal Technology-TOJET, vol. 11, no. 4, pp. 418–426,
2012.
[52] X. Zhai, Q. Fang, Y. Dong et al., The effects of
biofeedbackbased smulated recall on self-regulated
online learning: a gender and cognive taxonomy
perspecve, Journal of Computer Assisted Learning, vol.
34, no. 6, pp. 775–786, 2018.
[53] C. C. Chase, D. B. Chin, M. A. Oppezzo, and D. L. Schwartz,
Teachable agents and the proteg´ e effect: increasing the
effort´ towards learning,Journal of Science Educaon and
Technology, vol. 18, no. 4, pp. 334–352, 2009.
[54] B. M. McLaren, K. E. DeLeeuw, and R. E. Mayer, Polite
webbased intelligent tutors: can they improve learning in
classrooms?” Computers & Educaon, vol. 56, no. 3, pp.
574–584, 2011.
[55] G. Ngai, S. C. F. Chan, J. C. Y. Chan, and W. W. Y. Lau,
“Deploying a wearable compung plaorm for compung

Preview text:

lOMoAR cPSD| 59062190 Hindawi Complexity
Volume 2021, Ar cle ID 8812542, 18 pages h ps://doi.org/10.1155/2021/8812542 Review Ar cle
A Review of Artificial Intelligence (AI) in Education from 2010 to 2020 1 2 Xuesong Zhai , Xiaoyan Chu
,1 Ching Sing Chai , Morris Siu Yung Jong ,2 Andreja
Istenic ,3,4,5 Michael Spector ,6 Jia-Bao Liu ,7 Jing Yuan,8 and Yan Li 1
1 Zhejiang University, Hangzhou 310058, China 2
Chinese University of Hong Kong, Hong Kong 999077, Hong Kong 3
University of Primorska, Faculty of Educa on, Koper 6000, Slovenia
4 University of Ljubljana, Faculty of Civil and Geode c Engineering, Ljubljana 1000, Slovenia 5
Federal University of Kazan, Ins tute of Psychology and Educa on, Kazan 420008, Russia 6
University of North Texas, Denton 76207, USA
7 Anhui Jianzhu University, Hefei 230601, China 8
Anhui Xinhua University, Hefei 230088, China
Correspondence should be addressed to Yan Li; yanli@zju.edu.cn
Received 27 August 2020; Revised 18 January 2021; Accepted 2 April 2021; Published 20 April 2021 Academic Editor: Ning Cai
Copyright © 2021 Xuesong Zhai et al. This is an open access ar cle distributed under the Crea ve Commons A ribu on License,
which permits unrestricted use, distribu on, and reproduc on in any medium, provided the original work is properly cited.
This study provided a content analysis of studies aiming to disclose how ar ficial intel igence (AI) has been applied to the
educa on sector and explore the poten al research trends and chal enges of AI in educa on. A total of 100 papers including 63
empirical papers (74 studies) and 37 analy c papers were selected from the educa on and educa onal research category of
Social Sciences Cita on Index database from 2010 to 2020. The content analysis showed that the research ques ons could be
classified into development layer (classifica on, matching, recommenda on, and deep learning), applica on layer (feedback,
reasoning, and adap ve learning), and integra on layer (affec on compu ng, role-playing, immersive learning, and
gamifica on). Moreover, four research trends, including Internet of Things, swarm intel igence, deep learning, and neuroscience,
as wel as an assessment of AI in educa on, were suggested for further inves ga on. However, we also proposed the chal enges
in educa on may be caused by AI with regard to inappropriate use of AI techniques, changing roles of teachers and students, as
wel as social and ethical issues. The results provide insights into an overview of the AI used for educa on domain, which helps
to strengthen the theore cal founda on of AI in educa on and provides a promising channel for educators and AI engineers to
carry out further col abora ve research. 1. Introduction
the next disrup ve innova on [4]. AI is currently viewed by
many as a driver that is integral to the fourth industrial
The emergence of big data, cloud compu ng, ar ficial
revolu on, and it may trigger the fourth revolu on in
neural networks, and machine learning has enabled
educa on. Learning about AI has also begun to be part of
engineers to create a machine that can simulate human
school curriculum [5, 6]. However, just as the emergence of
intel igence. Building on these technologies, this study
television and computers was once touted to be game
refers to machines that are able to perceive, recognize,
changers of educa on, they have been shown to in fact
learn, react, and solve problems as ar ficial intel igence (AI)
enhance access to informa on without substan al y
[1, 2]. Inevitably, such smart technologies wil revolu onize
changing the core educa onal prac ces. Nonetheless,
the workplaces of the future [3]. Thus, while AI can interact
educators are obliged to review current AI capabili es and
and help humans perform at higher levels, it is emerging as
iden fy possible pathways to op mize learning. Given the lOMoAR cPSD| 59062190 2 Complexity
increasing a en on, it is mely to review recent AI research
model. This review focuses on ar cles published in the
in educa on to provide educators with an updated
period from 2010 to 2020 from the Web of Science, as that
understanding of the field as a prepara on to possible
represents the period when the second and third genera on changes.
of AI began to make headways into educa on. The research
AI has been increasingly propagated as having strategic
ques ons that guided this review are as fol ows:
value for educa on [7]. Loeckx [8] suggested that AI could
be an effec ve learning tool that lessens the burdens of
(1) What is the overal state of AI in educa on? Which
both teachers and students and offers effec ve learning
research topics and research designs related to AI in
experiences for students. Coupled with current educa on
educa on are evident from 2010 to 2020?
reforms such as the digitaliza on of educa onal resources,
(2) What are the trends in published studies in terms of
gamifica on, and personalized learning experiences, there AI in educa on?
are many opportuni es for the development of AI
(3) What are the chal enges generated from the current
applica ons in educa on. For example, the model ing research of AI in educa on?
poten al of AI techniques has been exploited systema cal y
to develop reac ve and adap ve tutorials for the 2. Method
construc on of individualized learning environments as
compensa on for the shortage of teachers through the use
This study is a systema c literature review. The objec ves of
of intel igent tutoring system (ITS) [10]. ITSs provide
the review were to analyze and interpret findings based on
personalized learning experience in four main ways:
predefined research ques ons (see above) and criteria
monitoring student’s input, delivering appropriate tasks,
which serve to point out future direc ons [15]. The
providing effec ve feedback, and applying interfaces for
predefined research foci as shown in Table 1 are research
human-computer communica on [7]. When more ITSs are
purpose, learning subject, educa onal level, research
created for more subjects and topics, it is likely to change
approach, and effects. The review was conducted in three
the role of teachers, and hence, schooling may need to be
stages: planning, performing, and repor ng the systema c
reconceptualized. There exist many concerns and worries review.
among teachers on if AI chal enges their jobs. At the same
me, such ques ons as what is being learned and how AI is
2.1. Planning the Review. As previous reviews about AI were
being used are being discussed currently by researchers as
conducted in the physical sciences [16, 17], the study aimed
wel as by educa onal prac oners. Some researchers
to conduct a review in the field of the social sciences.
wondered whether advancements in AI would chal enge or
The Web of Science database and the Social Science
even replace teachers since many other jobs are being
Cita on Index (SSCI) journals were selected for the search
replaced by automa on [11]. There is an emerging
for desired ar cles published from 2010 to 2020. Ar cles
recogni on that teachers’ professional roles need to be
published in the SSCI database are general y considered as
adjusted as AI advances and this wil trigger new
high-quality publica on among educa on researchers. The
organiza onal forms [12]. Emerging chal enges also
keyword employed was “ar ficial intel igence,” and the
included students’ a tudes towards these changes [13]. To
subject area was refined to “educa on and educa onal
some extent, students as digital ci zens are able to leverage
research”. This process yielded 142 ar cles including 121
AI to improve learning outcomes. Nonetheless, they may fail
research ar cles, 10 review papers, one interview paper,
to use suitable AI techniques appropriately for a specific
and 5 book reviews. The selected ar cles include both
learning context, which would result in nega ve a tudes
analy c studies (primarily qualita ve research) and towards learning [14].
empirical studies (primarily quan ta ve research).
To summarize, this research involves a review of the
studies of AI in educa on. Previous studies have included
three essen al perspec ves of AI in knowledge processing:
2.2. Performing the Review. Fol owing Wu et al. [18], this
(a) knowledge representa on, (b) knowledge obtaining, and
study was conducted in two steps: iden fica on and coding.
(c) knowledge deriva on [3]; this review wil focus on AI
In the first step, an ar cle was selected to the poten al pool
techniques and tools that have been integrated into
when it qualified for either of two criteria: (a) the research
educa on recently a er the prolifera on of AI. The “first
involved a specific AI technique as an interven on in
genera on” of AI could support human intel ectual work by
assis ng learning or teaching and (b) it provided empirical
applying rule-based expert knowledge, and the “second
evidence or in-depth analysis. As already noted, only ar cles
genera on” may find the op mal solu on by sta s cal/
indexed in SSCI were considered. It should be noted that
search model, while the “third genera on” wil drama cal y
studies that focused on the development processes of AI
improve recogni on performance based on the brain
without educa onal implica ons or only adopted AI as a
learning subject without the employment of AI were lOMoAR cPSD| 59062190 Complexity 3
excluded from this review. Second, as for the analy c
In the second step, al the authors discussed thema c
studies, only studies that discussed the effect of AI
analysis principles and established a coding scheme in terms
techniques on educa on were included. Each ful text of al
of how AI was used in educa on. Two main categories were
the iden fied papers was read and screened individual y by
inves gated: research ques ons and technology adop on.
three-panel members with doctoral degrees or
Firstly, with regard to research ques ons, previous research
professorships in the field of learning technology. Studies
has found three basic models of AI in knowledge processing:
that did not fit clearly with the criteria were brought up for
knowledge representa on, knowledge obtaining, and
panel discussion. The screening process yielded 100 ar cles
knowledge deriva on [3]. Building on that founda on, the
out of the original set of 121.
Table 1: The ar cles coded by research ques on, technology adop on, learning subject, educa onal level, research approach, and effects. ID Authors Research ques on Technology adop on Learning Subject Educa onal level Approach Effects 58 6th grade students (study 1); teachers and 1 Chin et al. [38] FEE Teachable agent Science educa on 134 EXP OUT∗ 5th grade students (study 2) 2 Ngai et al. [55] AFF Wearable biofeedback circuit Circuitry 7 to 9 grade students QE, SUR OUT∗ Intel igent 100 undergraduates and 3 Wegerif et al. [42] REA matchingpa ern Online dialogue 12 postgraduate students DA OTH+ algorithms 4 Thomas and Young [57] GAM Adap ve model ing Educa onal game 16 col ege graduates EXP, SUR PER∗ 5 Yang et al. [27] ADP Higher-order item response algorithm
Elementary mathema cs 158 six graders in Taiwan EXP PER+ 6 Moon et al. [58] GAM Experience point data model ing Digital game 40 plays EXP PER+ 7 McLaren et al. [54] AFF Intel igent tutoring system Chemistry 132 high school students QE OUT∗ 8 Jones [43]
Inves ga ng decentralized theory of ar ficial intel igence Exploring crea ve thinking 9 Va am et al. [24] REA Visualiza on Science educa on 157 middle school student QE OTH+ 10 Jonassen [45]
Introducing an ask system: interac ve learning system 11 Magnisalis et al. [21]
Review of adap ve and intel igent systems for col abora ve learning support: adap ve and intel igent systems Albin-Clark et al. ROL 4 early childhood lectures 12 [56] GAM Graphic simula on Construc on EXP, SUR PER+ and many students 13 Wong and Looi Exploring swarm intel igence [59] 14 Seni [60]
Inves ga ng the rela onship between neurosciences and organiza onal cogni on EXP, 15 Lin et al. [51] AFF Facial recogni on Digital art course 20 adults PER∗ SUR, INT 16 Heslep [61]
Introspec on to the misunderstandings of AI in educa on mo vated by AI enthusiasts About 500 Vietnamese news on many kinds of 17 Nguyen and Yang [28] MAT Extrac on algorithm Language learning DA 0 mobile phone from 2009 lOMoAR cPSD| 59062190 4 Complexity to 2010 Five interviews were Natural language conducted genera ng 18 Tierney [22] REC Language learning INT 0 process over seven hours of recordings 19 Lawler and Rushby [4]
Interview with Rover Lawler to give comments on the effect of computer technique on AI in educa on 20 Tufekçi and K¨ ose¨ [34] FEE Constraint-based model ing Programming
120 university students EXP, SUR OUT+ 21 Zipitria et al. [19] ADP Automa c discourse measure Language learning 17 summaries wri en in Basque language EXP PER+ 22 Chin et al. [39] REA Teachable agent Kit-based science curriculum 153 fourth grade students QE OTH∗ 12 pupils; 4 teachers; 2 23 Mukherjee et al. technical professionals; 2 QE OTH+ [30] FEE Text-to-diagram conversion Reading nontechnical persons Two undergraduate 24 Jain et al. [41] FEE Visualiza on History classes in computer QE PER+ science Game team (not 25 Higgins and men oned the 0 Heilman [62] REC Automated scoring system Language learning educa onal level) 148 students involved 26 Melo et al. [9] REC AFF Computa onal were either in high school EXP PER organiza on Mul disciplines ∗ or in early col ege years 27 Flogie and Aberˇsek [13] AFF Transdisciplinary pedagogy Natural science 100 students in 7th, 8th, 9th grades SUR OTH+ lOMoAR cPSD| 59062190 Complexity 5 Table 1: Con nued. Research ID Authors Technology adop on Learning Subject Educa onal level Approach Effects ques on 28 Rapanta and REA Argument map
Emira and Spanish classes 205 university students EXP OTH+ Walton [40] 476 mo on problems Visualiza on from 9th grade 29 Nabiyev et al. CLA Intel igent tutoring Mathema cal [29] mathema cs textbooks of DA 0 system Turkish Ministry of Educa on 30 Loeckx [8]
Analy c essay of opportuni es for AI used in educa onal data mining, adap ve learning, and crea vity Comparison of ar ficial networks, 31 Horakov´ a et al.´ classifica on, Ar ficial neural networks 120 text fragments QE, DA 0 [3] CLA regression trees, and decision trees 32 Ijaz et al. [14] IMM Virtual reality History 60 undergraduate university students QE PER+ 33 Liu et al. [31] REC Intel igent tutoring system Language learning 30 sports ar cles including 100 sentences DA 0 34 Malik and Ahmad [32] DEE E-assessment system Engineering 243 student of 8th graders EXP 0 16 ques ons from 35 Malik et al. [23] DEE Query trend Microso Students’ QA DA 0 assessment system Language learning Corpus K-means algorithm, 36 Peng [63] CLA REC PageRank algorithm Online learning More than 700 scholars EXP 0 37 MacIntyre et al. [64] MAT Text minding so ware Language learning 10 accomplished adult musicians and dancers INT 0 38 Aoun [65]
Book review in terms of importance and limita on of AI in educa on 39 Wil iamson et al.
Discussing the importance of neuroscience in educa on [33] Munawar et al. Intel igent virtual 40 [35] FEE
E-laboratory environment 161 university students SUR OTH∗ laboratory Learning system on
Telecommunica on 28 students studying diagnosis, 41 Samarakou et al. assistance, EXP OTH+ [47] ADP networks informa cs and evalua on 42 Fenwick [12]
Pondering the transforma on of teacher’ professional roles 43 Kessler [44]
Analy c essay of AI in the language teaching 44 Pe t et al. [36] GAM Online educa onal programming pla orm Programming 400 students and 12 teachers EXP OUT+ lOMoAR cPSD| 59062190 6 Complexity Table 1: Con nued. 8-9 grade A large archive database of text transcripts of 451 45 Kel y et al. [37] DEE Ques on authen city observa ons from 112 DA 0 measuring system
English and language arts classrooms and 132 high- quality audio recording from 27 classroom Students of 2016 from a col ege are selected for PE Autonomous learning 46 Ge et al. [20] CLA tes ng. Samples of the DA 0 MAT system Sports 150 ques ons are col ected 176 valid enterprise ques onnaires and 178 47 Sun [66] FEE Learning system Language learning
student ques onnaires are QE, SUR OTH∗ obtained 48 Auerbach et al. [67] IMM Robo c hardware and so ware pla orm Ar ficial Evolu on 42 postgraduate students EXP OUT∗ 49 Boulet and
Exploring online assessments system applied for the measurement in medical educa on Durning [68] Research ID Authors Technology adop on Learning Subject Educa onal level Approach Effects ques on
127 ques onnaires and 47 audio recordings from Ar ficial intel igence 50 Cukurova et al. Deba ng skil s SUR, DA OTH+ [69] REA and mul modal data candidates who have applied to become a tutor STEAM, conceptual Intel igent tutoring
(Not men oned the understanding, and 51 Du Boulay [70] ADP EXP OTH∗ systems (ITSs)
educa onal level) dialogue-based learning 52 Hughes [71]
Review on papers about early self- and coregula on from ar ficial intel igence perspec ve included Kay and Lifelong and life-wide (Not men oned the 53 Kummerfeld [72] FEE Learning system 0 personal user models educa onal level) 54 Ki o and Knight [73]
Inves ga ng three tensions in ethics when applying ar ficial intel igence and data analysis (AIDA) in educa on Data in one case from high 55 Luckin and Cukurova [74] REA
Learning sciences- Problem-solving, learning driven AI data, and deba ng school EXP,INT OUT∗ 4 semistructured interviews with five 56 Sel ar and Gulson senior policymakers, INT 0 [75] DEE AI and data science Educa on policy technical staff, and data scien sts
Online adap ve selfassessment procedure Thirty-two undergraduate 57 Sharma et al. [76] ADP
Web technologies with mul modal students EXP 0 data 58 Wang and Wang
Developing an ar ficial intel igence anxiety (AIA) scale [77]
Exploring the rela onships between AIA and mo vated learning behaviour lOMoAR cPSD| 59062190 Complexity 7 Table 1: Con nued.
59 Webb et al. [78] Discussing how me and temporality are used and inflected with the introduc on of AI in educa on policy contexts 60 Wil iams [79]
Analyzing implica ons of ar ficial intel igence, data analy cs, and blockchain technology for the academy Learning analy cs, AI, 61 Wil iamson [80] FEE and other so ware for Higher educa on Higher educa on SUR OUT+ data col ec on 62 Winters et al. [81]
Inves ga ng the exis ng digital structural violence and the approaches to tackling it 63 Rowe [82]
Exploring the effect on educa on reform brought by intangible economy which is shaped by globalized datasets
such as OECD PISA and ar ficial intel igence 64 Al y [83]
Iden fying the shaping forces for future educa on and competencies required by future digital teachers 65 Song and Wang [84]
Analyzing analysis of worldwide educa onal ar ficial intel igence research development in recent twenty years 30 students from a state 66 Ulum [85] FEE Versant English test English language learning DA,INT PER+ university in Turkey Educa onal data Mul linear regression 67 Costa-Mendes et al. [86] REA High school grades col ec on from preschool, EXP OUT+
model primary, and high school 68 Zhai et al. [87]
Inves ga ng the factors impac ng machine-human score agreements in machine learning-based science assessments First year students from 69 Lo us and FEE Bayesian networks Internet of Things Bachelor of Science in EXP OUT+ Madden [88] Compu ng program 70 Breines and Gal agher [89]
Introducing the applica on cases of teacherbot in the University of Edinburgh Middle, a Moodle plug- 71 Campo et al. [90] REA in using a Bayesian Computer science 45 university students EXP OUT∗ network model
Cri cal y reviewing the research on the use of social y assis ve robots (SARs) in the preter ary classroom and its 72 Papadopoulos et al. [91] benefits and disadvantages
73 Berendt et al. [92] Examining benefits and risks of ar ficial intel igence (AI) in educa on in rela on to fundamental human rights
Teachers selected learning material from a library to 67 par cipants aged 74 Standen et al. ADP The MaTHiSiS system create their own learning EXP OTH [93] ∗
between 6 and 18 years ac vi es and learning graphs Research ID Authors Technology adop on Learning Subject Educa onal level Approach Effects ques on 870 observa ons have been col ected from 5 75 Liu et al. [94] FEE BP neural network Undergraduate educa on EXP OUT+ consecu ve academic years in one university 76 Knox [6]
Analyzing the poli cal economy of ar ficial intel igence (AI) and educa on in China, with government policy and
private sector enterprise introduced
Students studying at the CGScholar (Common 77 Cope et al. [95] FEE Disciplinary knowledge masters and doctoral EXP OUT∗ Ground Scholar) levels lOMoAR cPSD| 59062190 8 Complexity Table 1: Con nued. onents an d their 78
Westera et al. Reviewing the ar ficial intel igence (AI) for serious games, presen ng reusable game AI comp [96]
relevance for learning and teaching, AI approach, and applica on cases Bonneton-Bo e´ The Kaligo, a digital 79 et al. [97] FEE Handwri ng Kindergarten EXP OUT∗ notebook applica on Smutny and (Not men oned the 80 Schreiberova [98] ROL Chatbots Disciplinary knowledge DA 0 educa onal level) Natural language (Not men oned the 81 Lucy et al. [99] DEE History EXP OTH∗ processing educa onal level) 82 Yakubu et al. [100] DEE
Ar ficial neural Learning management 1116 students in four network (ANN) systems (LMS) Nigerian universi es SUR PER∗
Analyzing th e educa on through 21st-century skil s and the impact of AI development in the age of pla orms, 83 Bonami et al. [101]
taking research, applica on, and evalua on into considera on Inquire Biology 24 students from 84 Koc-Januchta´ (ar ficial intel igence- Biology EXP,SUR OUT et al. [102] DEE ∗ Stockholm University enriched textbook) 85 Tran and
Introducing four innova ve projects that aim to extend learning management systems and improve the level of Meacheam [118] automa on 86 Nye et al. [103] DEE MentorPal STEM 31 high school students SUR PER∗
87 Webb et al. [104] Inves ga ng the implica ons of recent developments in machine learning for human learners and learning 3552 students from a 88 Tsai et al. [105] DEE Deep neural networks Disciplinary knowledge SUR 0 university in Taiwan 89 Alyahyan and
Construc ng guidelines to apply data mining techniques to predict student success Dustegor [106]
Analyzing th e drivers and barriers that currently affect data-based teaching and learning paths from the 90 Renz and Hilbig [107]
perspec ve of EdTech companies Gulson and
Inves ga ng how automated educa on governance assemblage includes new forms of exper se and authority 91 Witzenberger
and cons tutes EduTech as an important policy space [108]
Review the r esearch aimed at studying robot-man interac on, taking Russia and Kazakhstan as an example of the 92 Kerimbayev et al. [109]
interna onal coopera on in the sphere of robo cs 93 Fu et al. [110] FEE AI-enabled learning
Language learning 15 language learners INT,SUR PER+ tools
ning designs from the perspec ve of learning
94 Salas-Pilco [111] Examining t he use of ar ficial intel igence (AI) and robo cs in lear sciences
53 five-year-old and 49 seven-year- old Turkish 95 Yıldız [112] FEE Conceptualiza on EXP OUT∗ SCM-AI performances monolingual children from a primary school
96 Tolsgaard et al. Cri cal y rev iewing the published applica on and poten al role of data science and machine learning in Health [113] Professions Educa on 97 Hsu [114] DEE AI Chatbot English language learning 30 university students EXP OTH∗ 98 Wu et al. [115] CLA Machine learning A hybrid advanced sta s cs classifica on model
24 university students EXP 0 course 99 Wang et al. [116] ADP Squirrel AI learning Math 200 eighth grade students EXP OUT∗ lOMoAR cPSD| 59062190 Complexity 9 Table 1: Con nued. Rybinski and Natural language 640,349 reviews of 132 100 Kopciuszewska AFF processing (NLP) Higher educa on EXP 0 [117] models universi es
CLA: classifica on; MAT: matching; REC: recommenda on; DEE: deep learning; FEE: feedback; REA: reasoning; ADP: adap ve learning; AFF: affec on
compu ng; ROL: role-playing; IMM: immersive learning; GAM: gamifica on; EXP: experiment; QE: quasiexperiment; DA: discourse analysis; INT: interview;
SUR: survey; OUT: outcome; PER: percep on; OTH: others including affec on, cri cal thinking, and crea vity. ∗: sta s cal y significant change. +:
recognizable change without conduc ng significance tests. 0: focus on algorithms test without examina on of learning performance. lOMoAR cPSD| 59062190 10 Complexity
research ques ons of the sample papers were classified into
(ITS) and electronic assessment. The development
three dimensions: (a) development, focusing on the
procedure was usual y conducted with an induc on-
knowledge presenta on model; (b) extrac on, centering on
deduc on approach, in which prior experiments and data
how to obtain knowledge from data mining; and (c)
were analyzed to predict the variables fol owed by the
applica on, emphasizing the human-computer interac on
algorithm tes ng to obtain the final model ing equa on
through informa on deriva on. Secondly, with regard to [19].
technology adop on, the focus was on the types of
General y, the development of an educa onal system is
technology that the study adopted, which were further
cons tuted of three components: the presenta ons, logical
categorized into so ware (e.g., algorithms and programs)
model ing, and data dimension [20]. Al the 23 studies
and hardware (e.g., sensors and devices such as virtual
centered on logical model ing, while no study was found on
reality). It should be noted that a study with technology
the presenta on methods or data mining. The possible
without an AI purpose in educa on was not included. A
explana on may due to that the model ing techniques were
detailed descrip on is shown in Table 1 and it includes
the founda on of AI technique and fundamental y
learning subject, educa onal level, research approach, and
penetrate throughout the procedure of system
effects. Moreover, the researchers conducted further
development. In this dimension, the research was general y
frequency comparisons on the associa ons between the
conducted in the domain of computer science or
research purposes and some factors such as AI technology
informa on science, and the domain knowledge as the
adop on as wel as me periods to predict the trends and
source material was imported into algorithm frame (shown chal enges of AI in educa on.
in Figure 1(a)) with few pedagogical designs reported. For
example, Horakova et al. [3] aimed to explore the 3. Findings and Discussion
classifica on ability of a text mining machine using three
classifica on techniques. The results show that ar ficial
According to the above coding criteria and content analysis,
neural networks (ANNs) were significantly more effec ve
the three dimensions of research ques ons are shown in
than regression trees and decision trees to separate
Table 2 and the 72 studies from 63 empirical studies (5
educa onal texts or text fragments.
papers have two studies and 2 papers have three studies)
Addi onal y, in terms of the matching/group forma on
are further subclassified into 11 categories. There are 23
model ing, prior research employing stereotype theory has
studies in the dimension of development. The AI technique
assessed that the Bayesian networks, associa on rules,
was u lized as a development tool for the construc on of a
clustering, fuzzy C-means, and the fuzzy and gene c
smart learning environment, which can be subclassified as
algorithms were wel -accepted algorithms for the model ing
focusing on the development of algorithms including
of individual proper es of the student. These techniques
classifica on, matching, recommenda on, and deep
provide poten al indica ons for the inves ga on of forming
learning for teaching and learning purposes. Addi onal y, 35
homogeneous and heterogeneous groups in an educa onal
reviewed studies were found in the dimension of extrac on, context [21].
which referred to the applica on of developed AI
Moreover, the trends of the growing amount of data
techniques, normal y based on algorithms, to offer students
chal enge educators to analyze qualita ve data efficiently.
feedback, reasoning, and adap ve learning. 14 empirical
Natural language processing (NLP) provided a means to
studies were found in the dimension of applica on which
diagnose the problem and make a recommenda on by
consisted of affec on compu ng, role-playing, immersive
simplifying and accelera ng the discovery of what lies
learning, and gamifica on. In the integra on dimension, AI
within the data [22]. However, the assessment of a complex
techniques included those involving human factors as vital
educa onal system requires more profound informa on
variables to iden fy and analyze learners’ personalized
retrieval. The integra on of mul ple approaches, such as
features. In such studies, human-computer interac on was
benchmark in NLP/Seman c Web field, was suggested to
generated to improve such characteris cs as crea vity,
model smarter computer-aided systems in which agents
responsibility, and cri cal thinking that can impact learners’
could be trained automa cal y [23].
performances and percep ons. The fol owing sec ons
To op mize the model ing in the learning context, the
describe what educa onal issues were dealt with in the age
hierarchical structures were considered as poten al
of AI and how AI technique was employed in each research
solu ons to model the educa onal system. This is because ques on.
educa on is general y a complex system with the exhibi on
of subsystems and components, in which the invisible causal
processes among subsystem/component behaviours would
3.1. Dimension of Development. As shown in Table 2, 16
causal y affect each other [24]. It was suggested that
empirical studies were found focusing on the development
systema c model ing should analyze three dimensions in
of educa on systems such as intel igent tutoring system
the educa on context: learner’s varia on, learning lOMoAR cPSD| 59062190 Complexity 11
domains, and learning ac vi es [25, 26]. For example, some
Based on the above and Nguyen and Yang’s sugges on
researchers constructed the higher-order item response
[28], the aims of developing an AI-integrated system in
theory framework involving the overal ability at the first
educa on could be grouped into four types: classifica on (5
dimension and mul ple domain abili es at the second
studies), matching (3 studies), recommenda on (5 studies),
dimension, which has been wel adopted in the automa c
and deep learning (10 studies). (1) Classifica on refers to the problem-solving process [27].
reconstruc on of knowledge bases, in which the materials
Table 2: The number of studies concerning AI in educa on from 2010 to 2020. Quan ta ve research topics Classifica on 1 2 1 1 5 Matching 1 1 1 3 Development (N 23) Recommenda on 1 2 2 5 Deep learning 2 1 1 6 10 Feedback 1 1 2 2 3 7 16 Extrac on (N 24) Reasoning 1 1 1 1 4 2 10 Adap ve learning 1 1 1 4 2 9 Affec on compu ng 1 1 1 1 1 1 6 Role-playing 1 1 2 Applica on (N 12) Immersive learning 1 1 2 Gamifica on 1 2 1 4 Quan ta ve research 4 6 3 3 5 1 2 8 7 13 20 72 Qualita ve research 0 3 3 1 0 0 1 1 3 9 16 37 Total 4 9 6 4 5 1 3 9 10 22 36 109 matching Reasoning Adap ve learning (a) (b) lOMoAR cPSD| 59062190 12 Complexity (c)
Figure 1: The hierarchy of ar ficial intel igence in educa onal implementa on. (a) The dimension of system development, (b) the
dimension of extrac on, and (c) the dimension of applica on.
could be categorized according to varied characteris cs.
teaching. There are currently some AI applica ons that have
Classifica on demarcates knowledge content, which
achieved the integra on of technique, domain knowledge,
contributes to the accuracy of text analysis [3]. For example,
and pedagogical design. The three types of pedagogical
some researchers developed an ITS with the characteris cs
applica ons of AI iden fied in this review were feedback (16
of categorizing mo on problems, by which learners could
studies), reasoning (10 studies), and adap ve learning (9
easily access different types of mo on problems in
studies). While these applica ons could be interlinked, they
Mathema cs [29]. (2) Matching refers to a conversion
were categorized as such based on the classifica on
mechanism, in which varied sets of classifica on are
explicated by the authors of the reviewed ar cles.
connected to specific learning purpose. For example, a text-
to-diagram system was developed for blind students to link
3.2.1. Feedback. One of the chal enges impairing
geometry words to an underlying diagram on the Brail e
personalized learning is the inappropriate sequencing of
printout, which has been cer fied as an effec ve
contents. The restructuring of presenta on sequences is
teaching/learning tool at a Blind school [30]. (3) The
seeking a way to redefine the organiza on of knowledge
recommenda on is regarded as an intel igent authoring
according to the student’s reac on. In this situa on,
tool. With the support of the natural language process, it
feedback is an important approach to meet learners’
could automa cal y create new themes, theories, and
proximal learning pa erns [9]. Using an ar ficial neural
pedagogical contents as a response to learners’ feedback, to
network, the system provides immediate feedback
help teachers save me and effort [31]. It constructed a
according to students’ input to help them gradual y get
human-computer interac on and widely used to generate
access to the abstract concepts and perform prac cal
real- me and intel igent feedback according to learners’
exercises. Besides, researchers perceived a posi ve trend
input, which has been regarded as a reliable feature in
towards the system, which may a ribute to two
modern assessment system [32]. (4) Deep learning, or perspec ves.
machine learning, is a comprehensive approach of big data
processing and learning behaviour analysis. Based on the
(1) Based on Ohlsson’s theory, students can learn from
prolifera on of big data in educa on, such as learning or
the feedback generated as the result of an error [34]. In a
teaching behaviour, the system could self-adjust to meet
physical teaching environment, the teacher could interact
users’ dynamic requirements by upgrading its algorithms
with students immediately as difficul es arise. It is, [33].
however, difficult for such just-in- me interac on in an
online context. The situa on requires intel igent algorithms
To date, some studies have reported the lack of
to provide feedback automa cal y. For example, with the
significant impact on improving teaching. The chal enge was
help of pedagogical agent-based cogni ve architecture, the
largely a ributed to the weak pedagogical design and lack
intel igent virtual laboratory was developed to give
of appropriate assessment criteria [8]. Future research
appropriate feedback to students who encounter difficul es
should therefore be grounded in learning theories so that
in the laboratory [35]. Besides, a learning website, Jutge.org,
more acceptable, accessible, and efficacious AI can be an
was developed with the features of a rich and wel -
integral part of learners’ lives.
organized problem repository. The website provides instant
feedback and helps students to progressively solve
3.2. Dimension of Extrac on. Educators have begun to
problems and learn from their mistakes [36]. (2) Immediate
explore suitable applica ons of AI techniques in their
feedback promotes ac ve training in interac ve learning lOMoAR cPSD| 59062190 Complexity 13
environments that would benefit learner’s comprehension
the sociocultural theories of learning in designing AI to
diagnosis [19]. The previous study combined speech
support students’ reasoning, Va am et al. [24] reported that
recogni on, natural language processing, and machine
engaged learners could be er understand the mul ple
learning to measure the quality of classroom talk, in which
levels of organiza on in complex systems. Therefore,
new forms of interac on were created to provoke thoughts
students’ engagement is an essen al aspect to be
and further shape the effec ve interac on of the learning
considered for the design of a learning system that aims to
environment [37]. Another AI system used path traversal support reasoning.
algorithms to establish causal chains, by which students
The hierarchical reasoning generated by the intel igent
were provided with elaborated feedback and hints rather
system had beneficial effects on students’ learning. Firstly, it
than the correct answers. The learning-by-teaching context
may help learners to op mize the elucida on of the
was constructed by learners’ self-organiza on of
rela onships between the subcomponents of a par cular
interac ons and their interpreta on of feedback [38].
topic. In return, the intel igent reasoning system can be used
Although a large number of benefits were reported with
as a form of evalua on to assess if the student has captured
respect to automated feedback of domain knowledge, no
enough concepts for the given topic [41]. Secondly, the
research in this review had established the connec on to
system could provide an argumenta ve interac on which
pedagogical theories. Most of the authors in the
placed great significance in the construc on of col abora ve
development dimensions were from the computer science
learning atmosphere. It is because, as a result of peers’
domain, which leads to their focus on the presenta on of
reasoning, learners tend to externalize their arguments and
source data (domain knowledge) technical y without much
improve their premises. Jain et al. [41] combined visualized pedagogical considera on.
mapping tool with col abora on scripts. The design
successful y helped learners to analyze and evaluate
opposing posi ons on conten ous topics. General y,
3.2.2. AI-Supported Reasoning. The recursive feedback may
researchers regarded the reasoning visualiza on tools as
have the poten al to foster learners’ abili es to reason in
valuable scaffolds to develop learners’ cri cal thinking and
specific ways because the human-computer interac on is wri ng [40].
able to engender among the students a sense of
However, using AI techniques, including visualiza on
responsibility toward improving the construc on of
and hierarchical reasoning model ing, may be inadequate to
knowledge repository [39]. The reconstruc on of the
support reasoning. The four studies reviewed focused on
knowledge repository was seen as a process of using
the u liza on of model ing to support general reasoning,
model ing to realize pedagogical design as shown in Figure
while the reasoning model should be largely domain-
1(b). However, some researchers found that novices such as
specific [24, 39, 40, 42]. Moreover, there is an unresolved
students and preservice teachers showed minimal
chal enge in coding learners’ behaviours as far as AI-
understanding of the invisible causal behaviours in the
supported reasoning is concerned. The reasoning process
system compared to experts and experienced teachers [24].
may be more effec ve when learners’ personalized
Another research showed a similar conclusion: students
performance is considered. Although the visualized
were able to learn the relevant facts and pairwise rela ons,
reasoning tools could perform wel in a smal -scale group
while they may s l fail to reason with them very wel [39].
se ng, it is difficult to obtain adequate reasoning analysis
One possible explana on could be that reasoning is largely
of the data from a large popula on because the reasoning
invisible and it is difficult to induce the processes of
system fails to adjust itself automa cal y. Therefore, the
reasoning through the observa on of the behaviours. AI
requirements of dealing with increasingly large and diverse
techniques such as the visualiza on technique could be
data demand self-adap ve alterna ves [9].
applied to foster learners’ reasoning.
To help learners improve their reasoning, the graph
structure [29] and learners’ engagement [24] techniques
3.2.3. Adap ve Learning. Based on the new decentralized
have been studied. For the graph structure, intel igent
theories of AI and social cogni on, the apparent complexity
systems could be developed to make thinking visible. In a
of learners’ behaviour was largely a reflec on of the
sense, the simula on approach of the AI technique was
complexity of the learning environments. This prompted
employed to mimic thoughts tracking the reasoning visual y
educators to provide adap ve scaffolds for diversified
in real me. For example, the argument-mapping tools were
learning environments with various types of learners.
designed to assist learners with visualiza on of the premises
Different from the feedback system that offers stock
and conclusions of arguments. The findings showed that a
responses, the adap ve educa onal system is a forma ve
sequence of connected arguments was chained together for
and correc ve automated system that can adjust itself
learners to make an ul mate conclusion [40]. Drawing from
(target of interven on) to suit individual learners’ lOMoAR cPSD| 59062190 14 Complexity
characteris cs, needs, and preferences (pedagogical
were less no ced. Based on the maturity of biofeedback
objec ve) [43]. Although only three empirical studies were
technique, such as eyetracking and EEG, affec on
iden fied in this review, some researchers were very
compu ng was increasingly adopted to inves gate
posi ve to the future promo on of adap ve system in
students' internal mo va ons on learning, such as crea vity
teaching and learning. Technologies such as intel igent and responsibility [49, 50].
speech recogni on and automated wri ng evalua on [44]
According to the content analysis of the selected papers
have been tested with promising findings. In addi on, there
shown in Table 1, there are five typical AI techniques that
was substan al evidence showing that adap ve intel igence
supported affec on compu ng and analysis in the
enhances learning by automa cal y enabling learners to
educa on sector. They are complex algorithms,
locate and access proximal educa onal resources with
visualiza on, XR (virtual/augmented/mixed reality),
respect to naviga on and presenta on support [45].
wearable technique, and neuroscience. In many situa ons,
Previous research has emphasized that the design
they supported each other to construct a smart learning
dimension was a worth exploring alterna ve in the
environment and system. (1) Complex algorithms were
applica on of adap ve system [46]. To design successful
designed with considera on of human factors rather than
adap ve systems in educa on, curriculum designers and
the simple combina on of func onal blocks. From the
system designers have to leverage on to include the
perspec ve of human-computer interac on, the learners
model ing of the problem-solving process in the specific
should be treated as a knowledge creator rather than the
domain knowledge and the use of big data [21, 44]. Firstly,
receiver, which helps to generate posi ve affec on status.
the mechanism of the adap ve system connects learners’
From the perspec ve of presenta on modes, the tradi onal
prior domain knowledge and the evalua on of their current
declara ve statements in a computer system should be
domain performance to scaffold their problem-solving [47].
replaced by more diversified verbal presenta ons such as
In par cular, the pedagogical design is essen al in adap ve
dialogue, coaching, and generality. (2) Visualiza on was
intel igent context. It involves the selec on of adap ve
seen as an op mal method chosen for the solu on of
algorithms and considera ons about the compa bility of
complex concep on. One of the benefits of visualiza on is
the learning style and the intelligence suppor ve methods.
making complex knowledge entertaining, such as game-
In this sense, the assump on that AI would threaten the
based learning, in which learners’ mo va on wil be greatly
teachers’ posi on may be unfounded because of teachers’
generated. (3) XR including virtual/augmented/mixed
vital role as curriculum designers. Secondly, the adap ve
reality provides a highly simulated learning context, which
system is empowered by big data. Since the main feature of
may be chal enging to realize in physical classrooms. For
the adap ve learning system is personaliza on,
example, to help learners understand complex landforms in
accumula on of big data such as the range of diverse
geography, XR indulges students into a lively and crea ve
individual characteris cs and learning style and preferences
status. (4) The wearable technique, such as Google glasses,
is necessary for intel igent personaliza on to be realized.
helps to integrate learning ac vity into somatosensory
However, research on personaliza on in the context of the
moves. Although it was s l in an exploratory period, it has
adap ve system is limited to the users’ characteris cs
great poten al to advance domain knowledge in a prac cal
related to domain knowledge. The deeper internal
context in daily life. (5) Modern neuroscience exploits how
characters, such as human mental status and crea vity,
the brain works and this expands the research of learning to
were barely no ced and studied [21]. This however has vital
include the learners’ physiological state. Research in this
research poten al with the development of advanced AI
area would enrich understanding about individual varia ons
techniques such as biofeedback techniques.
and could provide addi onal avenues to match instruc on
with the most op mal guidance. 3.3. Dimension of Applica on
3.3.2. The Categories of the Applica on Dimension. With the
3.3.1. Technology Adop on in the Applica on Dimension.
supports of the above five AI techniques, four types of
The dimension of applica on highlights the importance of
learning models were generated with the applica on of
including human affec on in the applica on of AI in
affec on analysis, which was biofeedback (6 studies),
educa on. The latest research has indicated that affec on
roleplaying (2 studies), immersive learning (2 studies), and
had increasingly been reported to exert a significant gamifica on (4 studies).
influence on decision-making, percep on, and learning
Affec on compu ng refers to the analysis of human
[48]. Previous studies on the measurement of learning
emo ons and feelings captured by physical sensors and
performance only focused on two dimensions: learning
affec ve algorithms, which has gained much a en on in
outcomes (e.g., scoring and achievement) and percep ons
recent years. Affec on compu ng enhanced human-
(e.g., sa sfac on and acceptance), whereas other aspects lOMoAR cPSD| 59062190 Complexity 15
computer interac on. Based on the facial iden fica on,
learning tools were in the explora ve stage. Further
some researchers improved the intel igent tutoring system
inves ga ons in specific domains are eagerly needed.
by which students’ emo onal status was detected to give
Gamifica on has emerged as an important theore cal
them mely emo onal feedback [51]. Two essen al aspects
no on in the educa on sector. The most successful
are needed to op mize the affec on compu ng technique:
educa onal games ghtly integrate the pedagogical design,
first, teachers have to make mely appropriate instruc onal
domain knowledge, and affec on elements with gameplay.
adjustments according to learners’ affec ve status; second,
AI has assisted the integra on of the game and knowledge
comprehensive opera on of mul mode affec on sources as
domain, and the further poten al is making the game adapt
a single source is unlikely to provide accurate analysis of
to the learners’ behaviours and affec ons dynamical y [57].
affec on. For example, the eye-tracking technique could
One of the examples appropriately integra ng domain
capture learners’ eye fixa on to track the a ended area, but
knowledge with affec on is Minecra Edu. This is a historical
the reasons for the foci may be a ributed to different
simula on game where students can learn about historical
affec ons such as interest, anxiety, or even distrac on. An
figures and events or get insight into the spread of
addi onal source of data such as EEG could help to make a
epidemics. Learners could get access to historical events
more accurate assessment [52].
with authen c emo ons in the real- me interac on, and
Role-play is a learning method that inspires students to
the col ateral emo on would help them be er understand
ponder on problems with affec ons assuming varied roles.
the specific content knowledge [8].
Some algorithms were designed with the integra on of
Another example employed a game reward system as
roleplay into the pedagogical design, where students are
mo va onal mechanisms to promote voluntary and
taught by an intel igent agent rather than being taught by
proac ve learning. The results showed that the reward
the learning system [39]. Enlis ng role-play can enhance
system had a desirable fit with the pedagogical design, and
learners’ investment in their interac ons with computers.
the future educa onal algorithms might be er get
More than that, learners’ sense of responsibility was
associated with the field of ar ficial intel igence to mo vate
exerted towards the intel igent agent, which was consistent emergent learning [58].
with the research from Chase et al., demonstra ng that
students may work harder on behalf of their agents than
they would for themselves [53]. Addi onal y, to mo vate
3.4. The Results from Qualita ve Research. According to
students to act as a companion to an intel igent agent, the
selected qualita ve research (as shown in Table 3), the
politeness presenta on mode was employed in the
explora on of AI in educa on experienced a process from
intel igent tutoring systems, which was observed to benefit
theore cal research to a specific prac ce field, and at last
the needy students [54]. The future research of role-play
back to review. Simultaneously, qualita ve research also
may focus on gran ng access to students so that they could
provided support for the development of quan ta ve
customize their roles and target agents.
research throughout the whole process. Some theore cal
Immersive learning is an approach that enables students
studies were at the forefront. For example, in 2011 and
to customize scenes of characters engaging in ful -view
2012, qualita ve research on decentralized theory [43] and
learning se ngs. The enhancement of XR, 3D graphics, and
swarm intel igence [59] appeared, and then the real
wearable devices could promote the learning performance
ar ficial intel igence research began. AI algorithms were not
and these are strongly related to immersive affec on, which
very mature at the beginning while advanced intel igent
generated students’ academic performance and posi ve
algorithms are usual y based on big data technology, and
percep ons, such as excitement, enthusiasm, and crea vity.
they could constantly learn and improve in the massive
For example, learners could obtain a high degree of
data. The big data must be decentralized and group-
excitement in the immersive learning environment.
oriented. Therefore, we believe that the early theore cal
Immersive environment can also be coupled with immersive
research has played a significant suppor ng role. In 2019,
col abora on with gestures, emo ons, and nonverbal
researchers a ached more emphasis on the summary of
communica on [14]. Using immersive learning may also
previous studies and prospects for future development, and
reduce students’ sense of being in midated by complex
more considera on wil be given to the status quo, future,
topics and technical concepts when they expose to
and possible problems of AI in various sectors of educa on.
simulated technological and compu ng issues [55]. Most
importantly, many immersive learning tools encourage
4. The Research Trends of AI in Education
learners’ enthusiasm to create and change the
environments, which could foster crea vity [56]. However,
4.1. Technology Adop on of Internet of Things. The exis ng
few studies were found to consider domain knowledge as a
research mainly focused on the virtual online system, and
variable. The possible reason may be that many immersive
the Internet of Things (IoT) is less no ced. Learners’
biofeedback also needs to be explored in future educa onal lOMoAR cPSD| 59062190 16 Complexity
research. According to the reviewed papers, a majority of AI Neuroscience in educa on
technology in educa on focused on online informa on
technology or system (107 out of 109), such as intel igent
tutoring system, intel igent virtual laboratory, and
assessment system. Only one study [55] employed a
wearable circuit to examine learners’ biofeedback. This may
Educa onal data mining, adap ve learning, and crea vity
be a ributed to the fact that the intel igent online system is
wel established, easier to build on, and cost-effec ve.
However, to cater to diverse learning contents and varied
learning skil s, the IoT holds much promise. It may enhance Swarm intel igence
students’ spa al and mechanical understanding of physical (2) Neurosciences in Edu
(3) Misunderstandings of AI in Edu
construc on processes in science educa on. The IoT
technology can simulate brain func ons in physical context
to sense and understand human’s cogni ve behaviours, 2011
which apparently op mizes human cogni on and (1) Decentralized theory
performance in two qualita ve studies [33, 60]. Although no (2) Interac ve learning
empirical studies in the selected papers were found to test
the effect of IoT technique on educa on, the IoT with
affordable costs and wearable compu ng devices could be
a poten al area of future development of AI in educa on.
4.2.SwarmIntel igenceinEduca on. Swarm intel igence has
This is consistent with the Horizon report in 2019.
become a vital development direc on of AI, where the roles
Table 3: Qualita ve research topics.
of teachers and students wil be disrup vely changed.
According to the selected papers, the decentralized theory 2020
was firstly inves gated in educa on in 2011 [43], fol owed
(1) AI research development in recent twenty years
by the introduc on of swarm intel igence in educa on in
(2) Machine learning-based science assessments
2012 [59]. However, no empirical study has explored how
(3) Applica on cases of teacherbot
teachers and students meet the chal enges brought by
(4) Social y assis ve robots (SARs)
swarm intel igence. It is predicted that the fol owing two
(5) AIED in rela on to fundamental human rights
topics may become the research trends according to the
(6) Poli cal economy of AI and educa on
features of swarm intel igence. Firstly, swarm intel igence (7) AI for serious games
does not rely on centralized control of individual
(8) Impact of AI development in the age of pla orms
behaviours. In this situa on, learners change from
(9) Innova ve projects extending LMS
knowledge absorbers to creators. They ac vely constructed (10) Machine learning
knowledge by interfacing with the system in a variety of (11) Data mining techniques
contexts. Teachers’ “authori es” may be chal enged by a
(12) Data-based teaching and learning paths
group of experienced prac oners such as engineers and
(13) Automated educa on governance assemblage
farmers, and a col abora ve curriculum design would be (14) Robot-human interac on
constructed by swarm intel igence system [45]. Moreover,
(15) AI and robo cs in learning designs
swarm intel igence may change teachers’ du es from
(16) Data science and machine learning
knowledge transmission to knowledge organiza on.
Previous research has suggested the explora on of 2019
crowdfunding or crowdsourcing by teachers on educa on, (1) Assessment system
and how teachers perform their organizing ability in the
(2) Early self- and coregula on from AI perspec ve
future [5]. However, as Figure 2 presents, the inves ga on
(3) Ethical tension about applying AIDA in educa on
from teachers’ perspec ve is s l inadequate, which needs
(4) Ar ficial Intel igence Anxiety (AIA) Scale
further study. Secondly, swarm intel igence facilitated
(5) AI in educa on policy contexts
adap vity in dynamic or unstable environments. Swarm
(6) AI, data analy cs, and blockchain technology
agents usual y exchange informa on by leaving marks and
(7) Digital structural violence
observing the ac vi es of their peers. For example, the best (8) Intangible economy
solu on in the current moment may become unavailable in
(9) Future educa on and digital teachers
the next moment. Therefore, it is suggested to invest further
how AI performs dynamic recommenda on for students on 2018
different learning progress [59]. lOMoAR cPSD| 59062190 Complexity 17 4.
3.DeepLearningandNeurocomputa on. Deep
5. The Challenges AI Confronted in Education
learning or machine learning wil reshape the interac ons
between human beings and machines in the future. The
AI is a promising field that faces many technology
trends of human-computer interac on wil no longer be
bo lenecks. The chal enges would be more complex and
based on the perspec ve of machine opera on by a human.
intricate, especial y when they are connected to an
Instead, the machine can improve predic ons by learning
applica on in educa on. The chal enges this review
from big data without being specifical y programmed. Two
iden fies could be classified into three categories:
studies on deep learning were first men oned in the
technique, teachers and students, and social ethics.
selected papers in 2017 [23, 32]. In 2018, one empirical
Although AI techniques displayed and predicted smart
study [37] was published and it focused the deep learning
computa on in the educa on domain, they general y fail to
technology on the model ing of scoring-based data.
bring “added-value” to large-scale students because of the
However, the data based on human’s physical features were
concern of costs, and the mainstream is s l occupied by
less no ced. Based on the basis of neuroscien fic
“basic value” [38]. Specifical y, some researchers found that
understanding of the brain, Pearson and IBM have proposed
many AI techniques were designed for a general situa on
to inves gate neurocomputa on brain-based educa onal
that could not address the needs of a par cular domain,
technologies [33]. However, only two qualita ve studies
specific learning ac vi es, or teaching goals. This would
[33, 60] suggested the integra on of neuroscience and AI in
prevent the actualiza on of personalized learning
the educa on sector. Future research trends in integra ng experiences [8, 120].
brain func on with deep learning techniques to op mize
Another great chal enge reported in the Horizon report
human-computer interac on could be expected. It wil
in 2018 is the reconceptualiza on of the role of educators.
influence the applica on and integra on of AI in educa on,
Teachers’ a tudes towards AI have a significant influence
such as adap ve learning and role-play. This view has been
on the effec veness of using AI in educa on. Teachers may
reported in the Horizon report in 2018. Specifical y, the
swing from total resistance to overreliance. The former
report forecasts that adap ve learning techniques wil be
could arise from inadequate, inappropriate, irrelevant, or
further generalized in two to three years.
outdated professional development. The la er may be due
to teachers’ unrealis c expecta ons. These teachers may
focus too much on the emerging AI technologies rather than
4.4. Evalua on of AI in Educa on. Al empirical studies
learning itself [44]. Addi onal y, from the perspec ve of
reviewed presented the posi ve effects of AI techniques on
students, AI technique may provide smart and efficient tools
educa on (see Table 1). However, the interview and the
that cause students to avoid doing the knowledge
review paper have, respec vely, surfaced the chal enges or
processing work that teachers expect them to do. For
misunderstanding of AI in educa on [4, 21]. There is a need
example, the AI translators may offer ready-made
to ar culate a holis c evalua on criterion to measure the
il ustra ons, pronuncia on, fixed phrases, and even a serial
effec veness of AI in educa on. To ensure the validity and
of examples. Students are thus unwil ing to engage in the
reliability of the evalua on, a mul dimensional model
inquiry processes that facilitate deep learning.
should be adopted, which includes technique, pedagogical
The ethical issues brought by AI are also chal enging for
design, domain knowledge, and human factors. Woolf’s
both researchers and educa onal prac oners. It was clear
[119] Roadmap for Educa on Technology predicted that in
that AI has made great strides over the past few years,
the era of AI Educa onal Data Mining, the lifelong
mostly because of cheaper processing and the availability of
assessment of students’ knowledge, their progress, and the
data; however, individual student data may be exposed,
environments where they learn, as wel as the success and
shared, or used inappropriately. It is a constantly mindful
failure in teaching strategies, can be chronological y tracked.
chal enge that educators and AI engineers wil face when
Besides, current research is dispropor onately focused
considering how we access, evaluate, and share the big data
on specific educa onal contexts and a handful of variables.
and the results of data analysis [44, 65]. Another ethical
As shown in Figure 2, most research sampled students as
debate was conspicuously found in gamifica on that
par cipants, while teachers and professor prac oners
emphasis should be put on learning and tend to “suck the
were less no ced; addi onal y, most researchers considered
fun out” of games, or on gameplay “suck out the learning”
science, humanity, and social science as subjects, but less [57].
a en on was paid to sports, arts, and special educa on. For
example, only one study was found to develop text-to-
diagram conversion as a novel teaching aid for blind learners [30]. lOMoAR cPSD| 59062190 18 Complexity Col ege Others students students social sciences Educa onal level Subjects (a) (b)
Figure 2: The number of reviewed studies by educa onal level and subjects. 6. Conclusions
that has published 30 volumes were not considered. This
review therefore is limited only to SSCI ar cles. Addi onal y,
Given the rapid growth of AI, there is an urgent need to
the ini al search could be extended using more keywords
understand how educators can best u lize AI techniques for
such as adap ve learning and tutor system, which may lead
the academic success of students. This paper reviewed AI in
to the latest technical reports of AI in educa on that were
educa on research from 2010 to 2020. It is found that the
not included in this paper. Secondly, since the current
research to date could be classified into three dimensions:
review was not a empted to be inclusive but to provide a
the dimension of development including classifica on,
systema c overview of AI in educa on, the analysis in this
matching, recommenda on, and deep learning; the
review may provide a framework for future research
dimension of an extrac on involving feedback, reasoning,
integra on. For example, a more formal meta-analysis could
and adap ve learning; and the dimension of applica on
be conducted on selected empirical studies that reported
including affec on compu ng, role-playing, immersive
effect sizes to see what impact on learning AI might be
learning, and gamifica on. Moreover, based on the research
having. Besides, the future analysis could go back further in
ques ons and the related AI techniques, four research
me to see if there were changes about the me that AI 2.0
trends were iden fied. They are the Internet of Things,
started to make headways into educa on.
swarm intelligence, deep learning, and neuroscience, as
wel as an assessment of the effect of AI in educa on. The
chal enges of AI in educa on were also conspicuously seen Data Availability
in terms of technique perspec ve, teachers’ and students’
The content analysis data used to support the findings of
roles, and social ethical issues. These findings could be
this study are included within the ar cle.
valuable references for educa onal researchers, students,
and AI developers who plan to contribute to the relevant Conflicts of Interest
studies. Furthermore, it seems clear that educators need to
work with AI engineers to address the gaps between
The authors declare that they have no conflicts of interest technique and pedagogy.
to report regarding the present study.
7. Limitations and Future Study Acknowledgments
Although this review does propose some valuable trends
This research work was supported by the 2020 Humani es
and poten al research direc ons for AI in educa on, there
and Social Science Projects of the Ministry of Educa on
exist several limita ons. Firstly, the papers reviewed in this
(Grant ID: 20YJC880118), Na onal Science Funding of China
study were filtered from Social Science Cita on Index, while
(Grant ID: 61977057), 2019 Na onal Social Science Funding
other databases on natural science (e.g., SCOPUS and EI)
of China (19ZDA364), and the project of Informa za on
and sources (e.g., reports, news, conference papers, and
Capability in University Governance System, Chinese
patents) could be involved to offer a more comprehensive
Associa on of Higher Educa on, 2020 (Grant no.
overview in this field. For instance, ar cles from the 2020ZDWT18).
Interna onal Journal of Ar ficial Intel igence in Educa on lOMoAR cPSD| 59062190 Complexity 19 References
review,” Mechanical Systems and Signal Processing, vol. 108, pp. 33– 47, 2018.
[1] K. Kumar and G. S. M. Thakur, “Advanced applica ons of
[17] Z. Wang and R. S. Srinivasan, “A review of ar ficial
neural networks and ar ficial intel igence: A review,”
intel igence based building energy use predic on:
Interna onal journal of informa on technology and
contras ng the capabili es of single and ensemble
computer science, vol. 4, no. 6, pp. 57–68, 2012.
predic on models,” Renewable and Sustainable Energy
[2] J. M. Spector and D. J. Muraida, Automa ng Instruc onal
Reviews, vol. 75, pp. 796–808, 2017.
Design: Concepts and Issues, Educa onal Technology
[18] Y.-T. Wu, H.-T. Hou, F.-K. Hwang et al., “A review of
Publica ons, Englewood Cliffs, NJ, USA, 1993.
interven on studies on technology-assisted instruc on
[3] T. Horakova, M. Houska, and L. Domeova, “Classifica on of
from 2005–2010,” Journal of Educa onal Technology &
the educa onal texts styles with the methods of ar ficial
Society, vol. 16, no. 3, pp. 191–203, 2013.
intel igence,” Journal of Bal c Science Educa on, vol. 16,
[19] I. Zipitria, A. Arruarte, and J. Elorriaga, “Discourse measures no. 3, pp. 324–336, 2017.
for Basque summary grading,” Interac ve Learning
[4] R. W. Lawler and N. Rushby, “An interview with Robert
Environments, vol. 21, no. 6, pp. 528–547, 2013.
Lawler,” Bri sh Journal of Educa onal Technology, vol. 44,
[20] X. Ge, Y. Yin, and S. Feng, “Applica on research of computer no. 1, pp. 20–30, 2013.
ar ficial intel igence in col ege student sports autonomous
[5] Dai, C. S. Chai, P. Y. Lin et al., “Promo ng students’
learning,” Kuram Ve Uygulamada Egi m Bilimleri, vol. 18,
wel being by developing their readiness for the ar ficial no. 5, pp. 2143–2154, 2018.
intel igence age,” Sustainability, vol. 12, no. 16, pp. 1–15,
[21] I. Magnisalis, S. Demetriadis, and A. Karakostas, “Adap ve 2020.
and intel igent systems for col abora ve learning support: a
[6] J. Knox, “Ar ficial intel igence and educa on in China,”
review of the field,” IEEE Transac ons on Learning
Learning, Media and Technology, vol. 45, no. 3, pp. 1–14,
Technologies, vol. 4, no. 1, pp. 5–20, 2011. 2020.
[22] P. Tierney, “A qualita ve analysis framework using natural
[7] A. Seldon and O. Abidoye, The Fourth Educa on Revolu on,
language processing and graph theory,” The Interna onal
pp. 1–14, University of Buckingham Press, London, UK,
Review of Research in Open and Distributed Learning, vol. 2018.
13, no. 5, pp. 173–189, 2012.
[8] J. Loeckx, “Blurring boundaries in educa on: context and
[23] K. R. Malik, R. R. Mir, M. Farhan, T. Rafiq, and M. Aslam,
impact of MOOCs,” The Interna onal Review of Research in
“Student query trend assessment with seman cal
Open and Distributed Learning, vol. 17, no. 3, pp. 92–121,
annota on and ar ficial intel igent mul -agents,” Eurasia 2016.
Journal of Mathema cs, Science and Technology Educa on,
[9] F. R. Melo, E. L. Flores, S. D. Carvalho, R. A. G. Teixeira,ˆ L. F.
vol. 13, no. 7, pp. 3893–3917, 2017.
B. Loja, and R. de Sousa Gomide, “Computa onal
[24] S. Va am, A. K. Goel, S. Rugaber et al., “Understanding
organiza on of didac c contents for personalized virtual
complex natural systems by ar cula ng structure-
learning environments,” Computers & Educa on, vol. 79,
behaviorfunc on models,” Educa onal Technology & pp. 126–137, 2014.
Society, vol. 14, no. 1, pp. 66–81, 2011.
[10] B. Boulay, “Ar ficial intel igence as an effec ve classroom
[25] A. Casamayor, A. Amandi, and M. Campo, “Intel igent
assistant,” IEEE Intel igent Systems, vol. 31, no. 6, pp. 76–
assistance for teachers in col abora ve E-learning 81, 2016.
environments,” Computers & Educa on, vol. 53, no. 4, pp.
[11] M. Lacity and L. P. Wil cocks, Robo c Process Automa on 1147–1154, 2009.
and Risk Mi ga on: The Defini ve Guide, SB Publishing,
[26] A. Gogoulou, E. Gouli, and M. Grigoriadou, “Adap ng and Ashford, UK, 2017.
personalizing the communica on in a synchronous
[12] T. Fenwick, “Pondering purposes, propel ing forwards,”
communica on tool,” Journal of Computer Assisted
Studies in Con nuing Educa on, vol. 40, no. 3, pp. 367–380,
Learning, vol. 24, no. 3, pp. 203–216, 2008. 2018.
[27] C.-W. Yang, B.-C. Kuo, and C.-H. Liao, “A HO-IRT based
[13] A. Flogie and B. Aberˇsek, “Transdisciplinary approach of
diagnos c assessment system with constructed response
science, technology, engineering and mathema cs
items,” Turkish Online Journal of Educa onal
educa on,” Journal of Bal c Science Educa on, vol. 14, no.
TechnologyTOJET, vol. 10, no. 4, pp. 46–51, 2011. 6, pp. 779–790, 2015.
[28] B.-A. Nguyen and D.-L. Yang, “A semi-automa c approach
[14] K. Ijaz, A. Bogdanovych, and T. Trescak, “Virtual worlds vs
to construct Vietnamese ontology from online text,” The
books and videos in history educa on,” Interac ve
Interna onal Review of Research in Open and Distributed
Learning Environments, vol. 25, no. 7, pp. 904–929, 2017.
Learning, vol. 13, no. 5, pp. 148–172, 2012.
[15] B. Kitchenham, P. Pretorius, D. Budgen et al., “Systema c
[29] V. V. Nabiyev, U. Çakirog˘lu, H. Karal, A. K. Eru¨mit, and¨ A.
literature reviews in so ware engineering–a ter ary study,”
Çebi, “Applica on of graph theory in an intel igent tutoring
Informa on and so ware technology, vol. 52, no. 8, pp.
system for solving mathema cal word problems,” Eurasia 792–805, 2010.
Journal of Mathema cs, Science & Technology Educa on,
[16] R. Trescak, B. Yang, E. Zio, and X. Chen, “Ar ficial
vol. 12, no. 4, pp. 687–701, 2016.
intel igence for fault diagnosis of rota ng machinery: a lOMoAR cPSD| 59062190 20 Complexity
[30] A. Mukherjee, U. Garain, and A. Biswas, “Experimen ng
dialogues,” Computers & Educa on, vol. 54, no. 3, pp. 613–
with automa c text-to-diagram conversion: a novel 621, 2010.
teaching aid for the blind people,” Journal of Educa onal
[43] A. Jones, “Philosophical and socio-cogni ve founda ons for
Technology & Society, vol. 17, no. 3, pp. 40–53, 2014.
teaching in higher educa on through col abora ve
[31] M. Liu, V. Rus, and L. Liu, “Automa c Chinese factual
approaches to student learning,” Educa onal Philosophy
ques on genera on,” IEEE Transac ons on Learning
and Theory, vol. 43, no. 9, pp. 997–1011, 2011.
Technologies, vol. 10, no. 2, pp. 194–204, 2017.
[44] G. Kessler, “Technology and the future of language
[32] K. R. Malik and T. Ahmad, “E-assessment data compa bility
teaching,” Foreign Language Annals, vol. 51, no. 1, pp. 205–
resolu on methodology with bidirec onal data 218, 2018.
transforma on,” EURASIA Journal of Mathema cs, Science
[45] D. H. Jonassen, “Ask systems: interroga ve access to
& Technology Educa on, vol. 13, no. 7, pp. 3969–3991,
mul ple ways of thinking,” Educa onal Technology 2017.
Research and Development, vol. 59, no. 1, pp. 159–175,
[33] B. Wil iamson, J. Pyke , and S. Nemorin, “Biosocial spaces 2011.
and neurocomputa onal governance: brain-based and
[46] E. Walker, N. Rummel, and K. R. Koedinger, “Beyond explicit
brain-targeted technologies in educa on,” Discourse:
feedback: new direc ons in adap ve col abora ve learning
Studies in the Cultural Poli cs of Educa on, vol. 39, no. 2,
support,” in Proceedings of the 9th Interna onal pp. 258–275, 2018.
Conference on Computer Supported Col abora ve Learning,
[34] A. Tu¨fekçi and U. Ko¨se, “Development of an ar ficial
vol. 1, Maratea, Italy, June 2009.
intel igence based so ware system on teaching computer
[47] M. Samarakou, G. Tsaganou, and A. Papadakis, “An E-
programming and evalua on of the system,” Hace epe
learning system for extrac ng text comprehension and
Universitesi¨ Eg˘i m Faku¨ltesi Dergisi, vol. 28, no. 2, pp.
learning style characteris cs,” Journal of Educa onal 469–481, 2013.
Technology & Society, vol. 21, no. 1, pp. 126–136, 2018.
[35] S. Munawar, S. K. Toor, M. Aslam, and M. Hamid, “Move to
[48] M. B. Ammar, M. Neji, A. M. Alimi, and G. Gouarderes,
smart learning environment: exploratory research of
“The` affec ve tutoring system,” Expert Systems with
chal enges in computer laboratory and design intel igent
Applica ons, vol. 37, no. 4, pp. 3013–3023, 2010.
virtual laboratory for eLearning technology,” Eurasia
[49] I. Arroyo, D. G. Cooper, W. Burleson, B. P. Woolf, K. Muldner,
Journal of Mathema cs, Science and Technology Educa on,
and R. Christopherson, “Emo on sensors go to school,”
vol. 14, no. 5, pp. 1645–1662, 2018.
Ar ficial Intel igence in Educa on, vol. 200, pp. 17–24,
[36] J. Pe t, S. Roura, J. Carmona et al., “Jutge.org: 2009.
characteris cs and experiences,” IEEE Transac ons on
[50] K. Floyd, J. A. Hess, L. A. Miczo, K. K. Halone, A. C.
Learning Technologies, vol. 11, no. 3, pp. 321–333, 2018.
Mikkelson, and K. J. Tusing, “Human affec on exchange:
[37] S. Kel y, A. M. Olney, P. Donnel y, M. Nystrand, and S. K.
VIII. Further evidence of the benefits of expressed
D’Mel o, “Automa cal y measuring ques on authen city in
affec on,” Communica on Quarterly, vol. 53, no. 3, pp.
real-world classrooms,” Educa onal Researcher, vol. 47, 285–303, 2005. no. 7, pp. 451–464, 2018.
[51] H.-C. K. Lin, C.-H. Wang, C.-J. Chao, and M.-K. Chien,
[38] D. B. Chin, I. M. Dohmen, B. H. Cheng, M. A. Oppezzo, C. C.
“Employing textual and facial emo on recogni on to
Chase, and D. L. Schwartz, “Preparing students for future design
learning with teachable agents,” Educa onal Technology
an affec ve tutoring system,” Turkish Online Journal of
Research and Development, vol. 58, no. 6, pp. 649–669,
Educa onal Technology-TOJET, vol. 11, no. 4, pp. 418–426, 2010. 2012.
[39] D. B. Chin, I. M. Dohmen, and D. L. Schwartz, “Young
[52] X. Zhai, Q. Fang, Y. Dong et al., “The effects of
children can learn scien fic reasoning with teachable
biofeedbackbased s mulated recal on self-regulated
agents,” IEEE Transac ons on Learning Technologies, vol. 6,
online learning: a gender and cogni ve taxonomy no. 3, pp. 248–257, 2013.
perspec ve,” Journal of Computer Assisted Learning, vol.
[40] C. Rapanta and D. Walton, “The use of argument maps as
34, no. 6, pp. 775–786, 2018.
an assessment tool in higher educa on,” Interna onal
[53] C. C. Chase, D. B. Chin, M. A. Oppezzo, and D. L. Schwartz,
Journal of Educa onal Research, vol. 79, pp. 211–221,
“Teachable agents and the proteg´ e effect: increasing the 2016.
effort´ towards learning,” Journal of Science Educa on and [41] G. P. Jain, V. P. Gurupur, J.
Technology, vol. 18, no. 4, pp. 334–352, 2009. L. Schroeder, and
[54] B. M. McLaren, K. E. DeLeeuw, and R. E. Mayer, “Polite
E. D. Faulkenberry, “Ar ficial intel igence-based student
webbased intel igent tutors: can they improve learning in
learning evalua on: a concept map-based approach for
classrooms?” Computers & Educa on, vol. 56, no. 3, pp.
analyzing a student’s understanding of a topic,” IEEE 574–584, 2011.
Transac ons on Learning Technologies, vol. 7, no. 3, pp.
[55] G. Ngai, S. C. F. Chan, J. C. Y. Chan, and W. W. Y. Lau, 267–279, 2014.
“Deploying a wearable compu ng pla orm for compu ng
[42] R. Wegerif, B. M. McLaren, M. Chamrada et al., “Exploring
crea ve thinking in graphical y mediated synchronous