Vol. 1 n.

1 (2020)
The conditions and results of innovation in
educational models

Applications of Artificial Intelligence to higher education: possibilities, evidence,

and challenges
Francesc Pedró, UNESCO International Institute for Higher Education

Abstract

The range of applications of artificial intelligence (AI) to education is increasing ceaselessly,

although its generalization still seems far away. Despite the enormous opportunities that AI can offer
to support teaching and learning, the development of applications for higher education carries
numerous implications and also ethical risks. Against this context, this contribution aims to offer a
review of AI applications in higher education, taking as a starting point the heritage of research
developed in the last two decades. It explores the definitions of AI in education and the elements and
methods that AI applications could bring to higher education, discussing the challenges that emerge
and finally suggesting some conclusions.

Sintesi
La varietà delle applicazioni dell’intelligenza artificiale (AI) all’educazione sta aumentando
continuamente, sebbene la sua diffusione generalizzata sembri ancora lontana. Nonostante le
eccezionali opportunità che l’AI può offrire in favore di insegnamento e apprendimento, la sua
applicazione nell’istruzione superiore ha numerose implicazioni e anche rischi etici. Su questo
sfondo, il presente contributo ha l’obiettivo di presentare una rassegna delle applicazioni dell’AI
nell’istruzione superiore, partendo dalle ricerche sviluppate negli ultimi due decenni. Si esplorano le
definizioni di AI in ambito educativo, gli elementi e i metodi che le applicazioni dell’AI possono
apportare all’istruzione superiore, discutendo le sfide che ne emergono e suggerendo, infine, alcune
conclusioni.

Keywords: artificial intelligence, higher education, innovation, educational technology

Parole chiave: intelligenza artificiale, istruzione superiore, innovazione, tecnologia educativa

www.iulresearch.it
IUL Research | Open Journal of IUL University CC BY-NC-ND 4.0
www.iuline.it
 Introduction
The range of applications of artificial intelligence (AI) to education is increasing ceaselessly,
although its generalization still seems far away (Popenici & Kerr, 2017). Despite the enormous
opportunities that AI can offer to support teaching and learning, the development of applications for
higher education carries numerous implications and also ethical risks. For example, in times of post-
crisis budget cuts, administrators may be tempted to replace teaching with cost-effective automated
AI solutions. Faculty members, teaching assistants, educational advisors, and administrative staff
may fear that intelligent tutors, expert systems, and chat robots will take their jobs away from them,
perhaps not without reason. The application of AI, in particular to learning analytics, requires vast
amounts of data, including confidential information about students and teachers, which raises serious
privacy and data protection issues.
Against this context, this contribution aims to offer a review of AI applications in higher education,
taking as a starting point the heritage of research developed in the last two decades. It explores the
definitions of AI in education and the elements and methods that AI applications could bring to higher
education, discussing the challenges that emerge and finally suggesting some conclusions.

1. AI in education: origins and recent developments

The birth of AI dates back to 1956 when John McCarthy organized a two-month workshop at
Dartmouth College in the United States. In the workshop proposal, McCarthy first used the term
artificial intelligence (Russell & Norvig, 2010). To him comes the conjecture that every aspect of
learning or any other intelligence feature can be described so precisely that a machine can be designed
to simulate it. Therefore, it would be a matter of finding ways to make machines use language, form
abstractions and concepts, and solve the kinds of problems now reserved for humans and improve
themselves. On this basis, Baker and Smith (Baker, Smith, & Anissa, 2019) offer a broad definition
of AI as computers that perform cognitive tasks generally associated with human minds, in particular
learning and problem-solving. Thus, AI does not refer to a single technology but is used as an
umbrella term that describes a wide range of technologies and methods, such as machine learning,
natural language processing, data mining, neural networks, or a variety of algorithms.
The field of AI has its origins in computer science and engineering but is heavily influenced by
other disciplines. There is little agreement among AI researchers about a possible standard definition
and understanding of AI - and intelligence in general. In the specific field of higher education, the
truth is that many professors do not know its scope and, above all, what it consists of (Vinuesa et al.,
2020).
AI and machine learning are often mentioned as equivalent, but they are not. Machine learning is
an AI method for grading and profiling, supervised and unsupervised, predicting how likely a student
is of dropping out of a course or joining a program, or identifying problems in written assignments.
In this sense, machine learning has been defined as a subfield of artificial intelligence that includes
software capable of recognizing patterns, making predictions, and applying the newly discovered
patterns to situations that were not included or covered by its initial design (Popenici & Kerr, 2017).
It is often difficult to distinguish clearly between learning analytics (LA) and AI in education. An
important reason for this is that evidence of LA’s actual application, and its effect is only available in
a very fragmented way (Renz, Krishnaraja, & Gronau, 2020). The most common definition of LA is
the measurement, collection, analysis, and reporting of data about students and their contexts, to
understand and optimize learning and the environments in which learning occurs (Phil & George,
2011). In this way, LA is also considered the process of developing actionable knowledge through

62
the definition of problems and the application of statistical modulations and analyses against existing
or future simulated data, [allowing] institutions to experiment with the data to gain knowledge,
improve the student learning experience and student outcomes, and identify improvements in the
efficiency and effectiveness of delivery. Therefore, LA is only one promising component of AI in
education.
Another fundamental concept for AI is that of rational agents. An agent is an element that perceives
its environment through sensors and, after processing the information received rationally, acts on that
environment through actuators (Russell & Norvig, 2010). The vacuum robot is a simple form of
intelligent agent, but things become more complex and open when we think of an automated vehicle.
Learning analytics is a powerful resource for making informed decisions and achieving better
learning outcomes. Learning analytics draws on the availability of large amounts of data that can be
scrutinized to extract knowledge or develop intelligent tools useful for educational or administrative
tasks. Nevertheless, analyzing and making the most of the data is not an easy task. Advanced data
mining techniques are used to do this, which in turn rely on other fields such as big data technologies
to handle large volumes of data, self-learning algorithms that learn from the data, and visualization
tools for efficient communication with the people whose decisions are meant to help inform.
All these layers of software for intelligent data processing allow to obtain knowledge, detect
learning patterns, predict future situations, or give recommendations to optimize the available
resources. Analytics is also a crucial step in the development of future AI solutions that, with the help
of powerful libraries, including natural language recognition, language translation, and game theory,
will allow, for example, to create avatars that simulate the behavior of a virtual teacher for students
or a teacher’s assistant. The bright prospects for the future make it possible to anticipate an AI
ecosystem that can help overcome the various challenges linked to higher education quality and
equity, particularly when it is offered in hybrid or fully distance mode. While the future of AI
solutions is auspicious in the medium term, current solutions focus on making the most of data mining
and analysis technologies.

2. Possible areas of application

There seems to be a growing, shared understanding of the advantages of using AI in learning
environments are: a) to increase students’ learning (experiences and effectiveness) and their
motivation for learning, and therefore reduce student drop-out or inactivity and increase completion
of studies; and b) to provide customized and adaptable learning paths through specific objectives set
by the teacher or student to support the learning process. However, the use of AI outside Australia,
the UK, and the USA is still relatively rare (Ifenthaler & Yau, 2019).

2.1. A global perspective

The student’s higher education experience can be seen as a series of interdependent, overlapping,
but not necessarily sequential phases (Kriti Khare, Stewart, & Khare, 2018). This life cycle approach
is often used by administrators to manage students’ lives, as it distinguishes the critical elements of
the experience that enable the design and delivery of focused administrative services. The student life
cycle in higher education is defined as the student’s journey from first contact with an institution to
becoming an alumnus. A student’s ultimate goal is academic achievement accompanied by personal
development through academic experience. The academic success of students, however, depends on
a composite of all aspects of the student’s life. These other aspects include mental well-being and
support, social interactions, sports and physical health, and life balance, all of which contribute to

63
the student’s higher education career experience (Morgan, 2013). At least in theory, there is no area
of activity at an institution of higher education in which AI cannot potentially have a noticeable
impact (Zeide, 2019).
First, there is the institutional use. Universities, particularly those engaged entirely in distance
education, are increasingly dependent on algorithms for marketing to prospective students, estimating
class size, planning curriculum, and allocating resources such as financial aid and facilities. This leads
to another application of AI, student support, which is increasingly used in higher education
institutions. Schools use machine learning in student orientation. Some applications help students to
schedule the loading of their courses automatically. Others recommend courses, majors, and career
paths, as guidance counselors or career service offices traditionally do. These tools propose
recommendations drawing on how students with similar data profiles performed in the past.
Another area for the use of AI in supporting students is "just in time" financial aid. Institutions of
higher education can use student data to provide students with microloans or last-minute advances if
they need financial assistance to, for example, make it to the end of the semester and not drop out.
One of the most prominent ways in which predictive analysis is being used in student support is in
early warning systems, analyzing a wide range of data – academic, non-academic, operational – to
identify students who are at risk of failure or drop-out, or who have mental health problems. This use
shows some of the real potentials of artificial intelligence – large data can give editors a more holistic
view of the student’s situation.
Finally, universities can apply artificial intelligence to improve teaching and learning processes.
This involves creating systems that respond to the pace and progress of individual users. Educational
software evaluates student progress and recommends specific parts of a course for students to revisit
or additional resources to consult. These are often referred to as custom learning platforms.
These efficiencies are expected to lead to greater effectiveness, effective teaching, learning,
institutional decisions, and guidance. So this is yet another promise of AI: that it will show educators
things that they cannot assess or even imagine given the limitations of human cognition and the
difficulty of dealing with many different variables and a wide range of learners.
Alternatively, it is also possible to classify the uses of AI in higher education, according to the
end-user (Baker et al., 2019), i.e.:
(a) the student;
(b) the teacher; and
(c) the AI system itself.
Student-oriented AI tools are computer applications that students use to learn a subject, i.e.,
adaptive or personalized learning management systems. Teacher-oriented systems are used to support
the teacher and reduce his/her workload by automating tasks such as administration, assessment,
feedback, or plagiarism detection. AI tools also provide information about students’ learning progress
so that the teacher can proactively offer support and guidance when needed. System-oriented
applications are tools that provide information to administrators and managers at the institutional
level to facilitate their decision-making processes based on evidence of student behavior, courses and
programs.

64
 2.2. The advantages for instruction
More specifically, in the case of instruction, there seems to be some consensus about three broad
categories of AI applications in education that are already available today (Luckin, Holmes, Griffiths,
& Forcier, 2016), such as personalized tutors, intelligent support for collaborative learning; and
intelligent virtual reality.
Intelligent Tutoring Systems (ITS) can be used to simulate custom tutoring. Based on learning
models, algorithms, and neural networks, they can make decisions about a particular student’s
learning path and the content to be selected, provide cognitive scaffolding and assistance, involving
the student himself in the dialogue. ITS has enormous potential, especially in large-scale distance
higher education institutions, which offer modules with thousands of students, where personalized
human tutoring is impossible beyond relatively frequent individualized attention, solely for economic
reasons. A wide range of research shows that learning is a social exercise: interaction and
collaboration are at the heart of the learning process (Jonassen, Davidson, Collins, Campbell, & Haag,
1995). However, online collaboration does not just happen. It must be facilitated and regulated
(Salmon, 2003). AI can contribute to collaborative learning by supporting the formation of adaptive
groups, by facilitating online group interaction, or by summarizing discussions that can be used by a
human tutor to guide students toward course goals and objectives.
Finally, intelligent virtual reality can be used to engage and guide students in virtual reality
environments. Virtual agents can pretend to act as teachers, facilitators, or student peers, for example,
in virtual or remote laboratories (Perez et al., 2017). With the advancement of AI and the availability
of large volumes of student data whose analysis can be crucial for analyzing learning and improving
it, there is no doubt that a renaissance of assessment is taking place (Luckin et al., 2016). AI can
instantly provide feedback and assessment at the precise moment when this information may be
critical to the teacher, the student, or both. Instead of assessment being a specific activity, to which
time and effort must be devoted, AI can be incorporated into learning activities for ongoing analysis
of student achievement, from the perspective of what has traditionally been called formative
assessment. Thus, for example, algorithms have been used to predict that a student will fail a task or
drop out of a course with high accuracy with predictive success scores ranging from 75 to 95%
(Bahadir, 2016).

3. The contributions of empirical research

To what extent have these theoretical possibilities given rise to empirical research? A recent review
of research on the applications of AI to higher education (Richter, Juarros, Bond, & Gouverneur,
2019) has shown that, in reality, leaving aside the theoretical possibilities, it is possible to group
everything that has been developed so far around four main areas:
(a) adaptive systems and personalization;
(b) monitoring and evaluation;
(c) profiling and forecasting; and
(d) intelligent tutoring systems.
The following table shows the percentage of studies dedicated to each area.

65
 Area Percentage
Profiling and forecasting (admission 39%
decisions and course scheduling; drop-out
and retention; student models and academic
achievement)

Intelligent tutoring systems (teaching

course content; diagnosing strengths and
automated feedback; preserving learning 19%
materials; facilitating collaboration; the
teacher’s perspective)
Evaluation and assessment (automated
grading; feedback; assessment of student
understanding, commitment and academic 24%
integrity; teaching evaluation)
Adaptation and customization systems
(teaching course content; recommending
customized content; supporting teachers and
learning design; using academic data to 18%
monitor and guide students; representing
knowledge in concept maps)
Total 100%

TABLE 1 - APPLICATIONS OF AI TO HIGHER EDUCATION (ADAPTED FROM RICHTER ET AL., 2019)

3.1. Profile and prediction

The basis for many applications of AI are models or profiles of learners that can predict, for
example, the likelihood of a student dropping out of a course or being admitted to a program, in order
to provide timely support or provide feedback and guidance on content-related issues throughout the
learning cycle. Analysis, modeling, and prediction are an essential part of data mining in education
(Krishna, Kumar, & Sri, 2018). Accurate prediction of student performance can be critical to making
admissions decisions and providing better educational services (Chen & Do, 2014). In general,
research shows that admissions decisions can be predicted with a high level of accuracy, so an AI
solution could relieve administrative staff and allow them to focus on the more complicated cases.
On the other hand, studies on drop-out and retention aim to develop early warning systems to detect
students at risk in their first year or to predict the attrition of university students in general). Many
other studies are concerned with profiling students and modeling learning behavior to predict their
academic achievement at course level.

3.2. Intelligent tutoring systems

Research on student profiles is an essential basis for intelligent tutoring systems and adaptive
learning environments. More research is thus needed on the effectiveness of these systems. One of
the most relevant meta-analysis in this particular areas was published over 5 years ago: Steenbergen-

66
Hu and Cooper (Saiying Steenbergen-Hu & Harris Cooper, 2014) found that ITS had a moderate
effect on student learning and was less effective than human tutoring; but it outperformed all other
methods of instruction (such as face-to-face instruction, reading printed or digital text, or homework
assignment).
In general, these ITSs focus on providing learning content to students while supporting them by
giving adaptive feedback and clues to solve content-related questions and detecting students’
difficulties/errors in carrying out the proposed activities. Other possibilities, although with little
empirical research, are:
• diagnosing the strengths or gaps in students’ knowledge, and providing automated
feedback;
• the organization and presentation of learning materials based on the needs of the students;
• facilitating collaboration between students;
• the maximization of the teaching effort, with the aim of reducing the teacher’s workload.
Despite its enormous relevance, this is an area with hardly any empirical research. Monitoring
discussion forums provides another opportunity to improve the effectiveness of collaborative forums.
Teaching assistants (humans) are currently asked to check the forum at least once a day so that all
questions can be answered within 24 hours. The same rules apply to e-mails, and intelligent tutors
are encouraged to ask students to post a question on the forum if their answer would be useful to other
students. By associating a set of resources with each topic, it is possible to point out these resources
to students, just as a teaching assistant would have done (K. Khare & Lam, 2018). The algorithms
can be used to time the student response, and if a response is not found or takes too long, a mechanism
can be put in place to notify a teaching assistant. Once the assistant solves the question, the algorithm
can be trained further to answer the next time or to mark specific questions that should go to the
assistant directly. Going beyond answering the questions, reviewing the forum messages for student
understanding, using content analysis and text mining techniques can determine whether the coverage
of the discussion is as expected. Several studies (S. Steenbergen-Hu & H. Cooper, 2014) have
indicated that intelligent tutors may be as good or better than human tutors; however, the evidence is
not clear, and the experimental designs used to date do not provide an unequivocal answer.

3.3. Monitoring and evaluation

In general, studies show that AI applications can perform evaluation and assessment tasks with
very high levels of accuracy and efficiency. However, because of the need for human supervision,
they are more applicable to courses or programs with large numbers of students. Interestingly,
however, it has been found that the benefits of using algorithms that find patterns in student responses
lead to increased requests for review of assessments and a move away from merely measuring student
knowledge and skills through multiple-choice tests.
Automated assessments are mostly direct tabulations of multiple-choice questions, where there is
a predetermined correct answer. Assessments and scoring of less discrete answer sets, particularly
long essay questions, are not likely to be received as warmly by academics. However, AI is being
applied to the grading of both short and long essay types and shows considerable success. Work in
this area includes automatic feedback. Automatic scoring of short-answer questions has been much
studied (Zhang, Shah, & Chi, 2016), but automatic essay scoring is a still-growing field of research
aimed at scoring long essay questions. Chen, Breslow, and DeBoer (X. Chen, Breslow, & DeBoer,
2018) analyzed a mixed learning environment and the effects of immediate corrective feedback on
student behavior. They found that the feedback led to a reflective study, and higher performance was

67
predicted for students who used the corrective feedback function. Fruitful research on feedback
examines a battery of student tools, such as intelligent agents that provide students with cues or
guidance when they are confused or stuck in their work, and automatic learning techniques to generate
automatic feedback and help improve student writing.

3.4. Adaptation and customization systems

This area refers to adaptive systems that offer content, materials, and exercises customized
according to the students’ behavioral profile. In this sense, they can support teachers in the design of
learning and teaching by focusing on extracting academic information from students to perform
diagnostic tasks and help teachers to provide more proactive personal guidance or, in addition to that
task, facilitate performance assessment and personalized assistance and feedback.
Adaptive learning has been much touted over the years and was considered one of the first benefits
of online or computer-based learning. In adaptive learning, a program can judge when a new topic
needs to be introduced, or an older topic needs to be revisited by the student. The models used by
intelligent tutoring systems could be useful here to determine when a student has effectively learned
a concept and is ready to move on to the next (Lin & Chi, 2016). Task data and practice questions, as
well as response time, are often used to find the "learned" state and build a student model that
represents the students’ knowledge. These systems provide feedback, just-in-time guidance, and
explanations when students make mistakes. They track learning outcomes and can determine content
appropriate to the student’s level of difficulty (VanLehn, 2006). In this way, the students’ learning
experience is more focused on the most appropriate lessons. The use of games in education draws on
a similar pedagogical approach of discovering within a dependent and interrelated environment
objects or situations of knowledge. Polin (Polin, 2018) explains the characteristics of games that make
them learning support spaces. Lamb, Annetta, Firestone, and Etopio ((Lamb, Annetta, Firestone, &
Etopio, 2018) provide information on the type of games – Serious Games, Serious Educational Games
and Educational Simulations – that have the most significant impact on student cognition and
behavior.

4. The challenges
Given the possibilities offered, at least theoretically, by AI in higher education, and having also
examined the areas in which empirical research has made progress, it remains to identify the
fundamental challenges that would have to be faced in order for these possibilities to materialize.

4.1. Teaching staff

There are no indications that the adoption of artificial intelligence-based applications is beginning
to become widespread in higher education, at least for the time being, although the educational
technology industry has not yet ceased to produce new developments. Their fundamental flaw is that
instead of addressing the problems and issues that teachers face daily, they promote new ways of
organizing teaching that clash with dominant traditional practices, often without rigorous evaluations
to support the supposed benefits of the new solutions. Not surprisingly, teachers listen to what
vendors have to say, but they do not necessarily accept everything proposed to them. In this context,
some countries have already designed policies that support the efforts of the domestic educational
technology industry to promote innovation, to intensify efforts and modalities of qualification and
empowerment of teacher demand, while supporting their innovative practices and, finally, to explore

68
how AI can contribute to a more vibrant and more evidence-based higher education learning
environment.
Platforms that support learning analytics can use predictive algorithms to help teachers diagnose
and anticipate the learning difficulties faced by students and thus implement customized interventions
to respond to those difficulties. While predictive algorithms facilitate data analysis and interpretation,
these algorithms are not what make learning analytics systems powerful. Their effectiveness lies in
their usefulness and relevance for students and their teachers. Real-time data processing should result
in real-time feedback, enabling faster pedagogical intervention and individualized instruction.
Teachers must continue to play the leading role. Managers and teachers must have sufficient
autonomy to manage their respective institutions and courses based on the principle that they are
more familiar with their students’ needs. Automated testing only serves this autonomy if faculty and
administrators are empowered to manage the delivery of educational services in their respective
institutions. Otherwise, the application of any AI-driven tool can only lead to confusion and
contradictory, if not outright negative, results.
Therefore, teachers will remain on the frontline of higher education: those who say that AI can
replace teachers are misinformed. The arguments they make reduce university teaching to the
performance of exclusively cognitive and routine tasks, ignore research that underlines the
importance of a human tutor to support the learning process, and neglect the creative and socio-
emotional aspects of training, which go beyond the mere transmission of knowledge (Bali, 2017).
Furthermore, university professors will always decide how and when it would be appropriate to use
the tools supported by AI using their proverbial autonomy. The development of these AI-supported
tools and their integration into higher education programs should be a participatory process, designed
to "provide the support that teachers need – not the support that technologists or designers think they
need" (Luckin et al., 2016). That said, AI-enabled technologies offer opportunities to automate
particular routine and administrative tasks, such as classification and record-keeping, that teachers
are currently performing. Automation of such tasks can free up teachers’ time, allowing them to
devote more energy to the creative, empathetic, and inspirational aspects of the academic profession.
For an eventual generalization of the use of AI in higher education institutions, teacher training is
a key aspect in enabling teachers to use educational data to improve their teaching strategies and
methodologies. To be able to use AI-supported technologies effectively, teachers would also need to
assimilate new skills, specifically:
• a clear understanding of how AI-supported systems can facilitate learning can make sound
value judgments about new products and solutions offered to them;
• research and data analysis skills can interpret the data provided by AI-based systems,
interrogate data, and provide students with feedback drawing on the perceptions that
emerge from the data;
• new management skills to effectively manage the human and artificial intelligence
resources at their disposal;
• a critical perspective on the ways AI and digital technologies affect human lives and new
frameworks of computer thinking and digital skills for students to understand the power,
dangers, and possibilities of AI.
Teacher education programs in higher education institutions should, therefore, take account of
these new skills. However, not only teachers have to be prepared to understand and comprehend the
new technological possibilities that digital and AI education are unfolding. The history of innovations
in education depicts a lost land of promises because of a lack of understanding of the prevailing

69
pedagogical culture and how teachers work in universities. To design new educational possibilities,
AI developers have to dialogue with teachers, content designers, and interdisciplinary specialists.
There are currently two distinct communities of researchers, namely those focusing on learning
analytics (LA) and educational data mining (EDM). These two communities overlap in terms of
objectives and techniques but differ in the extent to which the MDE researchers, coming from the
intelligent tutoring community, work on cognition at a tiny scale. MDE methods are drawn from
various disciplines, including data mining, machine learning, psychometrics and statistics,
information visualization, and computer modeling. The field of learning analytics focuses more on
learning content management systems and their relationship to academic achievement. It does this by
combining institutional data, statistical analysis, and predictive modeling to identify which students
need help and how teachers can intervene to improve academic achievement.

4.2. Meaningful research for teaching

While it is reasonable to expect an increase in research on the application of AI in higher education
in the coming years, it is also worth recalling the difficulties the sector faces to take stock of
educational research in a meaningful way for practice and policymaking. The particular field of
educational technology research clearly demonstrates that what researchers declare as key research
issues is often not related to teachers’ actual needs. The potential of technology to transform education
has often been affirmed, although it is widely accepted that, for various reasons, this potential has not
yet been realized as expected. In examining how decisions are made about the use of technology in
education, it is shocking to realize how little is known about the effects of technology on the quality
of university education and, more specifically, what particular uses of technology can lead to
improved learning. The current development of AI in education seems to be another example of this
well-known phenomenon. This situation is far from optimal in the most impoverished and most
resource-constrained developing contexts, where technology-based reforms are being promoted as a
remedy for poor economic and social conditions. Since many national initiatives in this context often
emphasize the provision of access to technology as intrinsic value-added, as shown during the
COVID-19 crisis, there has not been much research on the actual impact on learning.
It is commonly said that the adoption of an innovation depends essentially on the perception by
the end-users of the benefits of implementing a new strategy in relation to what they are currently
using. Applying this principle to the specific case of technology in higher education, it can be
expected that AI supports the design of new strategies to engage students:
• learn better, for example, in a more personalized way;
• learn more, i.e., get better learning outcomes; and
• learn different things, that is, achieve learning goals that only technology can allow.
Research should identify the strategies that make this possible and, ultimately, the conditions
under which these strategies could be generalized. The question of feasibility is essential for learning
because there may be many strategies that could be incompatible with the current configuration of
universities and even the teaching profession.
Educational research, related to technology and in general, is a complicated task because of the
very nature of education and because contextual conditions limit its ability to provide results from
which generalizations can be drawn, thus affecting its capacity to contribute to the creation of
universally valid theories. In education, there is a pervasive problem of these so-called "ubiquitous
interactions"; that is, the large number of variables that are inter-related and increase the difficulty of
isolating impacts or combining the results of different studies. Since it is complicated to isolate the

70
influence of strategies, the socio-economic context, motivation, and interaction between all these
variables make it very difficult to arrive at universal recipes. As there are no universal solutions, it is
imperative that research also has a local dimension, in each university’s classrooms, recognizing
professors and students as actors and not as mere beneficiaries or users of previously packaged
technological solutions. There is no doubt that research has a role to play in elucidating the role that
technological solutions play in improving the quality of university education. However, the right
research questions need to be asked. Since educational phenomena are quite complex and
multifaceted, the right questions do not concern the use or non-use of AI in higher education, in
absolute terms, but which AI solutions can best be adapted to the evolving teaching and learning
needs of each teacher and student given their closer reality. The potential of AI in higher education
may seem very promising, but unless it is appropriately integrated into everyday teaching and
learning practices, its educational effects will never become visible.

4.3. Ethics and transparency in the collection, use, and dissemination of data
The ethical dilemmas that arise in the large-scale collection, production, analysis, and
dissemination of data on people are another important consideration when addressing AI’s potential.
However, it should be noted that trying to understand the ethical implications of new technologies is
by no means a new goal. Over the past three decades or so, academics and practitioners have sought
to define some form of information ethics that can be summarized in one question: What should an
ethical use of technology look like? The emergence of data science has shifted the discourse from
information ethics to data ethics. Experts have advanced the notion that "it is not the hardware that
causes ethical problems ... it is what the hardware does with the software and the data that represents
the source of our new difficulties" (Mittelstadt, Allo, Taddeo, Wachter, & Floridi, 2016).
How do these general concerns translate into the specific field of AI applications in higher
education? Probably around issues like the following.
• Access to higher education. More and more educational institutions are using automatic
learning algorithms to accept or reject students. Two potential problems that arise are: (1)
Lack of transparency. Some programs cannot easily explain why particular students are
accepted while others are rejected. Should every student have the right to understand those
reasons? (2) Unfair discrimination. When machine learning algorithms are trained with a
specific data set (say with students from a Western European country), the result may not be
directly applicable to students from other parts of the world. The initial database could be
biased towards a particular group and therefore discriminate unfairly when used in a different
group.
• Recommendations to individual students. If the recommendations are the result of automated
learning based on a large data set, the resulting recommendation may not be suitable for
students from a different target group. If the recommendations are based on the individual
student’s learning history, then this problem no longer exists.
• Concentration of personal data. If – as in the digital world – educational platforms are owned
by a few major worldwide players, two concerns arise: (1) the concentration of personal
information (of students and teachers), which could create a risk to privacy. Large amounts
of personal data are an attractive target for cybercriminals; (2) dominant platforms could forge
data monopolies by monopolizing the market on the ability to develop the best algorithms.
This would give them great power and increase concerns about the lack of transparency in
making decisions about individual students’ learning paths.

71
 • Accountability – What if the automated decisions that guide students in their learning process
turn out to be wrong? Who or what is ultimately responsible and accountable? The owner of
the platform? The assigned teacher? The algorithm?
• Impact on the job. If AI systems increasingly automate tasks that teachers usually perform,
what will happen to their work? AI systems can assess the student’s initial competency level,
guide her through the different stages of the course, applying collective intelligence combined
with individual experience, automatically evaluate test results, and even simulate, to some
extent, student-teacher interaction.
Data privacy and security arise almost immediately in debates about data ethics (Zeide, 2019). The
main challenge is to be able to use personal data while ensuring that individuals’ personally
identifiable information and privacy preferences are protected. Installing the necessary safeguards to
prevent data theft is also critical. While the growth of regulatory frameworks at global, regional, and
national levels on the protection of personal data certainly demonstrates a growing understanding of
the urgency of the issue, many of these frameworks still do not provide adequate protection to
citizens, both in policy and in practice.

Conclusions
Applications of AI to education have, from the beginning, faced many difficulties in growing
because education systems around the world are, by their very nature, resistant to technological
change. For several decades, AI has been part of a vision that promises to transform education by
creating tutorial systems that help to personalize learning. This promise is beginning to unfold as
today’s technology is being experimented with different models around the world, raising many
questions for education in general and higher education. The lack of longitudinal studies and, on the
contrary, the overwhelming presence of descriptive and experimental studies from the technological
point of view, as well as the prevalence of quantitative methods – especially the quasi-experimental
ones – in the empirical studies (Richter et al., 2019) shows that there is still substantial scope for
educators and university teachers themselves, whatever their discipline, to bring their perspective that
could have a great impact on higher education, for example, by adopting approaches such as design-
oriented research (Easterday, Rees Lewis, & Gerber, 2018). A recent systematic review of the
literature on personalization in education through technology also unveiled the predominance of
proposals that often used quantitative methods (Bartolomé, Castañeda, & Adell, 2018). It has also
been noted that there are very few application studies and impact studies in learning achievement
(Misiejuk & Wasson, 2017).
The overall consequences of AI application development in higher education cannot yet be clearly
anticipated, but it seems very likely that they will remain a significant issue in the next twenty years.
The applications developed so far offer enormous pedagogical opportunities for intelligent student
support systems and support student learning in personalized and adaptive learning environments.
This applies, in particular, to open and distance learning universities, where AI could help to
overcome the issue of providing access to higher education to an ever-increasing number of students.
On the other hand, it could also help them to offer flexible, interactive, and personalized learning
opportunities, for example, by relieving teachers of repetitive tasks such as grading large numbers of
assignments, so that they can focus on human accompaniment with learning empathy.
It is crucial to emphasize that educational technology is not just a matter of technology – the real
concerns must focus on the pedagogical, ethical, social, cultural, and economic dimensions of AI.
The danger, of course, lies in viewing data and coding as an absolute, rather than a relative, source of
guidance and support (Selwyn, 2016). Education is too complex to be reduced solely to data analysis

72
and algorithms. As is the case with digital technologies in general, data alone do not offer a clear
technical solution to the dilemmas of education, no matter how shocking the results of the analysis
of large masses of data may be. In China, systems are currently being used to monitor students’
participation and expressions by recognizing their faces in the classroom (the so-called Intelligent
Campus Classroom Behavior Management System) and showing them to the teacher on a tablet. This
is an example of educational monitoring, and it is highly questionable whether such systems provide
real added value to a good teacher, who should be able to capture the dynamics in a learning group
(online and in a campus environment) and respond to their needs in an empathetic and pedagogically
meaningful way, as a result of his or her expert professional judgment. Therefore, it is crucial to adopt
an ethical perspective to start thinking about how the potential of algorithmic decision-making
systems incorporated in AI applications is being explored (Prinsloo, 2017). Besides, we must always
remember that AI systems, above all, require the control of humans over whose understanding of the
problems and their alternative solutions the AI applications are programmed (Kaplan & Haenlein,
2019). Some critical voices remind us that we must go beyond tools, and talk again about pedagogy,
as well as recognize the relevance of the human aspects of the use of digital technology in education
(Castañeda & Selwyn, 2018).
A recent UNESCO report on the challenges and opportunities of AI for sustainable development
deals with various areas that have important pedagogical, social and ethical dimensions, for example,
ensuring inclusion and equity in AI, preparing teachers who will be decisive in the development of
AI applications in education, setting up quality and inclusive data systems, or ethics and transparency
in the collection, use and dissemination of data (Pedró, Subosa, Rivas, & Valverde, 2019). There
remains a dramatic lack of critical reflection on the pedagogical and ethical implications and the risks
of applying AI applications in higher education. Privacy issues are rarely addressed in empirical
research on AI applications to higher education, as if they were not with them, as a recent systematic
review on the topic of learning analytics has shown (Misiejuk & Wasson, 2017)).
More research is needed by teachers and instructional designers on how to integrate AI
applications throughout the student life cycle to take advantage of the enormous opportunities they
offer to create intelligent learning and teaching systems. The limited presence of AI researchers in
university education departments is evidence of the need for educational perspectives on these
technological developments. The lack of theory could be a widespread syndrome in the field of
educational technology in general. A recent study has shown that over 40% of articles in three leading
educational technology journals were entirely a-theoretical (Hew, Lan, Tang, Jia, & Lo, 2019), in line
with other previous studies (Bartolomé et al., 2018). Most of the research focuses merely on analyzing
and searching for patterns in the data to develop models, make predictions that inform students and
teachers facing applications, or support management decisions using mathematical theories and
learning methods developed decades ago (Russell & Norvig, 2010). This type of research is now
possible thanks to the growth of computing power and the full availability of large digital student
data. Nevertheless, there is very little evidence for the advancement of pedagogical and psychological
theories of learning related to AI-driven educational technology. Researchers should, therefore, be
encouraged to be explicit about the theories underlying empirical studies on the development and
implementation of AI projects in order to broaden research at a broader level, helping to facilitate
understanding of the reasons for and mechanisms of this dynamic development that is set to have a
growing impact on higher education institutions worldwide.

73
 References

BAHADIR, E. (2016). Using Neural Network and Logistic Regression Analysis to Predict
Prospective Mathematics Teachers’ Academic Success upon Entering Graduate Education.
Educational Sciences: Theory and Practice, 16(3), 943–964.

BAKER, T., SMITH, L., & ANISSA, N. (2019). Educ-AI-tion rebooted? Exploring the future of
artificial intelligence in schools and colleges. https://www.nesta.org.uk/report/education-rebooted/

BALI, M. (2017). Against the 3A’s of EdTech: AI, Analytics, and Adaptive Technologies in
Education’. The Chronicle of Higher Education, 2018-04.
https://www.chronicle.com/blogs/profhacker/against-the-3as-of-edtech-ai-analytics-and-adaptive-
technologies-in-education/64604

BARTOLOMÉ, A., CASTAÑEDA, L., & ADELL, J. (2018). Personalisation in educational technology:
the absence of underlying pedagogies. International Journal of Educational Technology in Higher
Education, 15(1), 14. https://doi.org/10.1186/s41239-018-0095-0

CASTAÑEDA, L., & SELWYN, N. (2018). More than tools? Making sense of the ongoing digitizations
of higher education. International Journal of Educational Technology in Higher Education, 15(22).
https://doi.org/10.1186/s41239-018-0109-y

CHEN, J. F., & DO, Q. H. (2014). Training neural networks to predict student academic
performance: A comparison of cuckoo search and gravitational search algorithms. International
Journal of Computational Intelligence and Applications, 13(01).
https://doi.org/10.1142/S1469026814500059

CHEN, X., BRESLOW, L., & DEBOER, J. (2018). Analyzing productive learning behaviors for
students using immediate corrective feedback in a blended learning environment. Computers &
Education, (117), 59–74.

EASTERDAY, M. W., REES LEWIS, D. G., & GERBER, E. M. (2018). The logic of design research.
Learning: Research and Practice, 4(2), 131–160.

HEW, K. F., LAN, M., TANG, Y., JIA, C., & LO, C. K. (2019). Where is the “theory” within the field
of educational technology research? British Journal of Educational Technology, 50(3), 956–971.

IFENTHALER, D., & YAU, J. (2019). Higher Education Stakeholders’ Views on Learning Analytics
Policy Recommendations for Supporting Study Success. International Journal of Artificial
Intelligence, 1(1), 28–42.

JONASSEN, D., DAVIDSON, M., COLLINS, M., CAMPBELL, J., & HAAG, B. B. (1995). Constructivism
and computer‐mediated communication in distance education. American journal of distance
education, 9(2), 7–26.

KAPLAN, A., & HAENLEIN, M. (2019). Siri, Siri, in my hand: Who’s the fairest in the land? On the
interpretations, illustrations, and implications of artificial intelligence. Business Horizons, 62(1), 15–
25.

74
 KHARE, K., & LAM, H. (2018). Educational Data Mining (EDM): Researching impact on online
business education. In A. KHARE & D. HURST (Eds.), On the line: Business education in the digital
age. Springer, Cham.

KHARE, K., STEWART, B., & KHARE, A. (2018). Artificial Intelligence and the Student Experience:
An Institutional Perspective. IAFOR Journal of Education, 6(3), 63–78.

KRISHNA, K. P., KUMAR, M. M., & SRI, P. A. (2018). Student information system and performance
retrieval through dashboard. International Journal of Engineering & Technology, 7(2.7), 682–685.

LAMB, R. L., ANNETTA, L., FIRESTONE, J., & ETOPIO, E. (2018). A meta-analysis with examination
of moderators of student cognition, affect, and learning outcomes while using serious educational
games, serious games, and simulations. Computers in Human Behavior, (80), 158–167.

LIN, C., & CHI, M. (2016, June). Intervention-BKT: Incorporating instructional interventions into
Bayesian knowledge tracing. In A. MICARELLI, J. STAMPER, & K. PANOURGIA (Eds.), International
Conference on Intelligent Tutoring Systems (pp. 208–218). Springer, Cham.

LUCKIN, R., HOLMES, W., GRIFFITHS, M., & FORCIER, L. B. (2016). Intelligence unleashed: An
argument for AI in education. Pearson.

MISIEJUK, K., & WASSON, B. (2017). State of the field report on learning analytics.

MITTELSTADT, B. D., ALLO, P., TADDEO, M., WACHTER, S., & FLORIDI, L. (2016). The ethics of
algorithms: Mapping the debate. Big Data & Society, 3(2).
https://doi.org/10.1177/2053951716679679

MORGAN, M. (2013). The Student Experience Practitioner Model. In M. Morgan (Ed.), Improving
the student experience: A practical guide for universities and colleges (pp. 69–88). Routledge.

PEDRÓ, F., SUBOSA, M., RIVAS, A., & VALVERDE, P. (2019). Artificial intelligence in education:
challenges and opportunities for sustainable development.

PEREZ, S., MASSEY-ALLARD, J., BUTLER, D., IVES, J., BONN, D., YEE, N., & ROLL, I. (2017).
Identifying productive inquiry in virtual labs using sequence mining. In International Conference on
Artificial Intelligence in Education (pp. 287–298). Springer, Cham.

PHIL, L., & GEORGE, S. (2011). Penetrating the fog: Analytics in learning and education.
EDUCAUSE Review, 46(5), 31–40.

POLIN, L. G. (2018). A constructivist perspective on games in education. In D. W. KRITT (Ed.),

Constructivist education in an age of accountability (pp. 163–188). Palgrave Macmillan, Cham.

POPENICI, S. A., & KERR, S. (2017). Exploring the impact of artificial intelligence on teaching and
learning in higher education. Research and Practice in Technology Enhanced Learning, 12(1).
https://doi.org/10.1186/s41039-017-0062-8

75
 PRINSLOO, P. (2017). Fleeing from Frankenstein’s monster and meeting Kafka on the way:
Algorithmic decision-making in higher education. E-Learning and Digital Media, 14(3), 138–163.

RENZ, A., KRISHNARAJA, S., & GRONAU, E. (2020). Demystification of Artificial Intelligence in
Education. How much AI is really in the Educational Technology? International Journal of Artificial
Intelligence, 2(1), 14–29.

RICHTER, O. Z., JUARROS, V. M., BOND, M., & GOUVERNEUR, F. (2019). Systematic review of
research on artificial intelligence applications in higher education: where are the educators?
International Journal of Educational Technology in Higher Education(16).
https://doi.org/10.1186/s41239-019-0171-0

RUSSELL, S., & NORVIG, P. (2010). The Nature of Environments. Artificial Intelligence: A Modern
Approach (3rd ed.). Pearson Education.

SALMON, G. (2003). E-moderating: The key to teaching and learning online. Psychology Press.

SELWYN, N. (2016). Is technology good for education?. John Wiley & Sons.

STEENBERGEN-HU, S., & COOPER, H. (2014). A meta-analysis of the effectiveness of intelligent

tutoring systems on college students’ academic learning. Journal of Educational Psychology, 106(2),
331–347.

STEENBERGEN-HU, S., & COOPER, H. (2014). A meta-analysis of the effectiveness of intelligent

tutoring systems on college students’ academic learning. Journal of Educational Psychology, 106(2),
331.

VANLEHN, K. (2006). The behavior of tutoring systems. International Journal of Artificial

Intelligence in Education, 16(3), 227–265.

VINUESA, R., AZIZPOUR, H., LEITE, I., BALAAM, M., DIGNUM, V., DOMISCH, S., & NERINI, F. F.
(2020). The role of artificial intelligence in achieving the Sustainable Development Goals. Nature
Communications, 11(1), 1–10.

ZEIDE, E. (2019). Artificial Intelligence in Higher Education: Applications, Promise and Perils,
and Ethical Questions. EDUCAUSE Review (Summer), 31–39.

ZHANG, Y., SHAH, R., & CHI, M. (2016). Deep learning + student modeling + clustering: A recipe
for effective automatic short answer grading. In T. BARNES, M. CHI, & M. FENG (Eds.), Proceedings
of the 9th International Conference on Educational Data Mining (pp. 562–567). Educational Data
Mining Society.

76