The Teacher as a Facilitator for Learning
Flipped Classroom in a Master’s Course on Artificial Intelligence
Robin T. Bye
Software and Intelligent Control Engineering Laboratory, Department of ICT and Natural Sciences,
Faculty of Information Technology and Electrical Engineering, Norwegian University of Science and Technology
NTNU, Postboks 1517, NO-6025 Ålesund, Norway
Flipped Classroom, E-Learning, Active Learning, Constructive Alignment, Problem-Solving, CodinGame,
edX, C-4 Dynamite for Learning.
In this paper, I present a flipped classroom approach for teaching a master’s course on artificial intelligence.
Traditional lectures in the classroom are outsourced to an open online course to free up valuable time for active,
in-class learning activities. In addition, students design and implement intelligent algorithms for solving
a variety of relevant problems cherrypicked from online game-like code development platforms. Learning
activities are carefully chosen to align with intended learning outcomes, course curriculum, and assessment to
allow for learning to be constructed by the students themselves under guidance by the teacher, much in accord
with the theory of constructive alignment. Thus, the teacher acts as a facilitator for learning, much similar
to that of a personal trainer or a coach. I present an overview of relevant literature, the course content and
teaching methods, and a recent course evaluation, before I discuss some limiting frame factors and challenges
with the approach and point to future work.
The flipped, or inverted, classroom teaching method-
olody can perhaps most simply be defined as swap-
ping learning activities that traditionally have taken
place in-class with learning activities that tradition-
ally have taken place out-of-class (Lage et al., 2000).
However, as noted in a survey of research on flipped
classroom by Bishop and Verleger (2013), flipped
classroom constitutes much more than a mere reorder-
ing of activities performed at home or in class. They
give a more useful definition as the flipped classroom
being “an educational technique that consists of two
parts: interactive group learning activities inside the
classroom, and direct computer-based individual in-
struction outside the classroom. Elaborating on this
definition, Bishop and Verleger (2013) highlights hu-
man interaction as the key component of the in-class
activities, with a foundation in student-centred learn-
ing theories such as those of Piaget (1968) and Vy-
gotsky (1978), whereas explicit instruction methods
based on teacher-centred learning theories are out-
sourced to automated computer technology.
Similarly, Abeysekera and Dawson (2015) em-
phasise that in a flipped classroom, “the information-
transmission component of a traditional face-to-face
lecture is moved out of class time [. . . and replaced
with] active, collaborative tasks. Consequently, stu-
dents prepare for class by outside-class activities nor-
mally covered by traditional lectures, thus freeing up
valuable in-class time to student-centred learning ac-
tivities. After class, students can continue working on
in-class tasks they did not finish, explore some topics
in more detail, revise material, and further consolidate
knowledge (Abeysekera and Dawson, 2015).
1.1 Active Learning
According to Sotto (2007), higher education is domi-
nated by the transmission method of teaching, which
can be popularly rephrased as teaching by telling.
Synthesising research on the effectiveness of tradi-
tional lectures, Bligh (1998) shows that they are not
very effective for developing skills, values or personal
development, all of which are natural learning goals
in higher education. Flipped classroom moves this
one-way passive learning activity outside the class-
room and replaces it with active learning, which can
be defined as any teaching method that engages the
students in the learning process (Prince, 2004). How-
ever, pointed out by Bishop and Verleger (2013), this
definition could in principle also include lectures
Bye, R.
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence.
DOI: 10.5220/0006378601840195
In Proceedings of the 9th International Conference on Computer Supported Education (CSEDU 2017) - Volume 1, pages 184-195
ISBN: 978-989-758-239-4
Copyright © 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
where students reflect, take notes, and ask questions.
Thus, to maintain a contrast with teacher-centred
learning, it can be useful to separate lecture-based
methods from traditional lectures and define active
learning as students participating in the learning pro-
cess, doing more than just passive listening (Bon-
well and Eison, 1991). In line with this, active learn-
ing is indeed an inseparable part of flipped classroom
(Schaathun and Schaathun, 2016).
Bishop and Verleger (2013) identify several ac-
tive learning paradigms, such as constructivism and
collaborative learning (originating from the theory
of cognitive conflict by Piaget (1968)), cooperative
learning (based on the theory of Vygotsky (1978)
on the zone of proximal development), peer-assisted
learning (defined by Topping and Ehly (1998) as
“the acquisition of knowledge and skill through ac-
tive helping and supporting among status equals or
matched companions”), and peer-tutoring (e.g., see
Topping, 1996, for a review). Another active learning
method is problem-based learning, which has over-
lap with learning methods under the umbrella of peer-
assisted learning but can also be done individually
(Bishop and Verleger, 2013).
Several metastudies show that active learning in
science, technology, engineering, and mathematics
(STEM) has several advantages regarding perfor-
mance, ability to reproduce material, and motiva-
tion and engagement (e.g., Prince, 2004; Schroeder
et al., 2007; Freeman et al., 2014), whilst cooperative
learning strategies have been shown to be particularly
effective for achieving deep learning (e.g., Foldnes,
2016; Bowen, 2000; Johnson et al., 1998; Springer
et al., 1999).
1.2 Constructive Alignment
For the last decades, there has been a dramatic change
in higher education worldwide, with many more stu-
dents enrolling, from a wider diversity of background,
and with a broader range of approaches to learning
(Biggs and Tang, 2011). In engineering, students
of today conceive learning as anything between sim-
ple memorisation of definitions, applying equations
and procedures, or making sense of physical con-
cepts and procedures, to seeing phenomena in the
world in a new way or changing as a person (Marshall
et al., 1999). The big variation among students in
this taxonomy ranging from surface learning (mem-
orisation) to deep learning (learning associated with
understanding and ultimately changing as a person),
with an added category of strategic learning, in which
students aim for good grades with minimal effort (En-
twistle and Ramsden, 1983), has necessarily had an
impact on how higher education is being taught
(Felder and Brent, 2005).
Because it is well documented that students’ ap-
proaches to learning has a significant effect on achiev-
ing learning outcomes (e.g., Gynnild, 2001; Marton,
1981), many studies have tried to identify factors that
promote deep learning (e.g., Marton and Booth, 1997;
Prosser and Trigwell, 1999). A popular answer for
dealing with the challenges of today’s diverse stu-
dent mass is the theory of constructive alignment
(Biggs, 1996). Constructive alignment (CA) merges
the constructivist view that students learn by doing
with aligning the teacher, the students, the teaching
context, the learning activities, and the learning out-
comes (Biggs and Tang, 2011). This is achieved by
working backwards when designing a course, starting
with the intended learning outcomes (ILOs), defin-
ing assessment tasks closely related to the ILOs, and
finally choosing teaching methods and learning ac-
tivities aligned with the ILOs and assessment tasks
(Biggs and Tang, 2011).
Biggs (e.g., Biggs, 1996, 2011; Biggs and Tang,
2011) is especially concerned about the assessment
tasks. Since many students are strategic in their ap-
proach to learning, the exam (assessment tasks) ef-
fectively defines the curriculum, as only those parts
of the course that will affect their final grade will be
prioritised. The answer in CA is to incorporate as-
sessment tasks with grades well aligned with ILOs to
ensure students achieve the latter.
1.3 Cognitive Load Theory
As noted in the literature review by Schaathun and
Schaathun (2016), active learning adopts the con-
structivist view that learners must construct their own
knowledge. However, at the same time, in order to
move new information from a working memory with
very limited capacity into long-term memory, deep
cognitive processing is required (Anderson, 2015).
This process, referred to as cognitive load theory
(Clark et al., 2011), represents a bottleneck in learning
when to complex problems and information must be
processed in too short a time. A remedy suggested by
several authors (e.g., see Sotto, 2007) is to break prob-
lems into small, manageable exercises in order to re-
duce the cognitive load. In flipped classroom, where
typically online video lectures are offered outside the
classroom, students can manage this cognitive load by
regulating pace and rewatching video lectures (Abey-
sekera and Dawson, 2015), as well as completing their
other learning tasks in their own order of preference,
with possible repetitions if necessary.
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence
1.4 Facilitating Learning
According to Biggs (2011), the teacher must create
a learning environment that facilitates learning activ-
ities that in turn make the students achieve the de-
sired learning outcomes. In CA, the key to success
is to make sure that all teaching and learning compo-
nents such as the curriculum and the ILOs, the teach-
ing methods, and the assessment tasks are aligned to
each other. However, it is commonly accepted that
lack of self-monitoring and self-regulation among
students will lead to poor academic results (e.g., Lan,
1996; Borkowski and Thorpe, 1994). Consequently,
the learning environment itself is not sufficient to
achieve ILOs, since students’ individual skill in self-
monitoring and self-regulation, that is, selecting and
structuring the material to be learnt, will highly af-
fect the learning outcomes (Gynnild et al., 2008). To
cope with this challenge, (Gynnild et al., 2007) sug-
gests that the teacher must adopt a role as a facilita-
tor for learning, much similar to a personal trainer at
the gym, guiding the trainee to do the right exercises,
adjusting the “weights, or cognitive load, whilst en-
couraging and supporting the trainee, and eventually
making the trainee self-monitored and self-regulated.
Obviously, this is a bold ambition, and increas-
ingly so with large classes, however, flipped class-
room may just be the tool that makes it possible.
1.5 Outline
In the following sections, I first give an overview of a
master’s course on artificial intelligence (AI) in which
I have employed flipped classroom. Next, I present
the teaching methods I have used and present a course
evaluation that includes feedback from students in the
2016 cohort. Finally, I discuss some limiting frame
factors and challenges to my approach, point to future
work, and draw some conclusions.
The elective third-semester master’s course IE502014
Topics in Artificial Intelligence (TAI) has been run
with two different cohorts (autumn 2015 and 2016)
since the inauguration in autumn 2014 of a new mas-
ter of science programme in simulation and visual-
ization at NTNU in Ålesund. The course consists of
7.5 ECTS
credits and runs for 14 weeks, where all
in-class activities are constrained to a single weekly
teaching day, referred to as workshops. Providing an
European Credit Transfer and Accumulation System.
introduction to the field of AI and a number of se-
lected topics therein relevant for solving real-world
problems, students study AI with respect to
modelling a variety of problems in suitable state space
design and implementation of intelligent search, opti-
mization, and classification algorithms
simulation and testing of models and algorithms
visualisation and analysis of the results
The ILOs of TAI are defined within the three cate-
gories of knowledge, skills, and general competence.
Specifically, upon completion of the course, students
should be able to
describe AI as the analysis and design of intelligent
agents or systems
explain terms such as perception, planning, learning,
and action as fundamental concepts in AI
define problems in suitable state space depending on
choice of solution method
solve problems by means of search methods, evolution-
ary algorithms, swarm algorithms, neural networks, or
other AI methods
collect information from scientific publications and
textbooks and reformulate the problem, choice of meth-
ods, and results in a short, concise manner
discuss and communicate advantages and limitations of
AI as a science
The course is split in two parts provided by two
teachers. Part A, taught by me, focusses on solving
problems by means of intelligent agents and the use
of various search algorithms, dealing with both unin-
formed and informed (heuristic) search, local search,
and adversarial search, whereas Part B, taught by my
colleague Associate Professor Ibrahim A. Hameed,
focusses on optimisation and constraint satisfaction
problems, as well as machine learning topics such
as fuzzy expert systems, artificial neural networks
(ANN), and hybrid intelligent systems. In relation to
this paper, the teaching methodology I apply for Part
A is the most relevant but I may refer to Part B or the
course as a whole where appropriate.
TAI has the following main components:
course textbooks and scientific literature
the learning management system (LMS) Fronter
three mandatory assignments
final oral exam
online code development platforms
massively open online courses (MOOCs)
full-day in-class workshops
In the following, I will present each of these compo-
nents in turn.
CSEDU 2017 - 9th International Conference on Computer Supported Education
2.1 Textbooks and Topics
The course relies heavily on the following textbooks:
Artificial Intelligence: A Modern Approach
(AIMA) (Russell and Norvig, 2013),
Artificial Intelligence: A Guide to Intelligent Systems
(AIGIS) (Negnevitsky, 2005), and
Learning for Data: A Short Course
(LFD) (Abu-Mostafa et al., 2012).
Both times we have run the course, the following ten
topics in the textbooks were studied (topics 1–5 con-
stitute Part A, topics 6–10 constitute Part B):
1. Introduction to AI (AIMA Ch. 1)
2. Intelligent Agents (AIMA Ch. 2)
3. Solving Problems by Searching (AIMA Ch. 3)
4. Beyond Classical Search (AIMA Ch. 4)
5. Adversarial Search (AIMA Ch. 5)
6. Constraint Satisfaction Problems (AIMA Ch. 6)
7. Fuzzy Expert Systems
(AIGIS Ch. 2–4, case study Ch. 9)
8. Artificial Neural Networks for Pattern Recognition
(AIGIS, Ch. 6, case studies Ch. 9)
9. Hybrid Intelligent Systems
(AIGIS Ch. 8, case studies Ch. 9)
10. Nonlinear Transformation and Feature Extraction
(LFD Ch. 3)
2.2 Fronter
is the official LMS in use at NTNU in Åle-
sund. Every course has its own “room” in Fronter that
enrolled students can access. The course room for
TAI contains a Frontpage with a summary of all rele-
vant information that is continously being posted, in-
cluding a welcome message; information about where
to buy the course textbooks; and course material
(course overview, oral exam information, assignment
handouts, and workshop material). The Frontpage
also contains internal hyperlinks to the course mate-
rial (usually pdfs) contained in a well-structured Doc-
uments directory. Students can quickly and easily ac-
cess the material from the Frontpage, which is impor-
tant if using a smartphone, but can also choose to nav-
igate the Documents directory for complete access to
all material. Assignments are handed in through the
Fronter submission system and detailed feedback on
students’ performance is also posted there. Using the
Fronter news functionality, we inform students when-
ever new material is available or when already posted
material has been updated.
The teachers have tried to encourage the Fron-
ter forum functionality by posting answers to several
questions that students ask in class or by email but
there has been zero activity from students in the fo-
rum. This is likely due to students being on-campus
and seeing each other steadily, thus reducing the need
for an online forum. Still, the teachers find the forum
functionality useful but then acting more as a knowl-
edge bank of frequently asked questions (FAQs) with
2.3 Assignments and Exam
Three mandatory assignments (two for Part A, one
for Part B) must be passed for permission to en-
ter the end-of-semester final oral exam. The assign-
ments typically consist of both theoretical short an-
swer questions and project-like programming exer-
cises. Employing a grading scheme consisting of the
letters A to F, where A is best and F is fail, a pass
grade is awarded if the work has the quality of a C
or better. However, the grade on assignments does
not contribute to the final grade, which is determined
solely by a final individual oral exam. Students usu-
ally get 3–4 weeks to complete and submit their as-
signments on Fronter.
Individual assignment reports, as well as a laptop
with computer code, can be brought to the exam to
provide individual entry points of discussion. The
oral exam is 25 mins and students are examined by
the course teachers.
2.4 Code Development Platforms
Developing a framework with simulation environ-
ments in which to test intelligent algorithms can
be a time-consuming task and divert attention from
the learning outcomes that we want the students to
achieve. For the programming exercises for Part A, I
rely on two online websites that provide such frame-
works, namely HackerRank
and CodinGame.
dents have to register a user (free of charge) on both
Using HackerRank and CodinGame removes the
need for writing a lot of boilerplate code and sim-
ulation logic, leaving the students to focus on the
design and implementation of the algorithms. Both
websites support more than 20 of the most common
programming languages, provide discussion forums,
blogs, and leaderboards (rankings), with CodinGame
also providing entertaining gamelike visualisations of
the code execution, which is quite useful when test-
ing an algorithm. Thus both websites provide fun and
playful environments that act as a catalyst for learning
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence
where students can compete both against themselves
and others in the quest of obtaining better scores.
2.4.1 HackerRank
HackerRank provides numerous programming chal-
lenges in several domains and programming lan-
guages, including the domain most relevant for this
course, namely AI. All challenges consist of a prob-
lem description and a code window in a web browser
where students can enter their programmes, or they
can use a plugin to enable writing their code locally in
their own editor. Programmes are then tested on text
input and must produce the correct text output. Sam-
ple code to get students started is provided in many
For Part A of TAI, I cherrypick exercises from
relevant AI subdomains on HackerRank, including
Graph Theory, A* Search, Bot Building, Alpha Beta
Pruning, Combinatorial Search, and Games. An ex-
ample exercise is BotClean Large, in which students
must program an intelligent agent (bot) to clean all the
dirty cells in a 2D grid. After submitting their code, a
simulation tests the bot of each student on a number
of different scenarios and returns a score.
2.4.2 CodinGame
CodinGame works in a similar manner as Hacker-
Rank but also provides game-like 2D graphical feed-
back of the behaviour of the code. The stub code
to get students started on a particular programming
challenge is usually better than the ones on Hack-
erRank. Programming challenges are divided into
single-player challenges (ranked Easy, Medium, or
Hard) that can be found in the Training category,
and larger, multiplayer games called Bot Program-
ming, that can be found in the Multiplayer category
(past contents) or Contests (currently running con-
tests). One example of a multiplayer game is Game
of Drones, in which students must control a fleet
of drones and compete against other players (class-
mates, online players, or the default AI provided by
CodinGame) with their own fleet of drones in a battle
of controlling as many zones as possible in a large 2D
2.5 Video Lectures and E-learning
With the advent of MOOCs, numerous online e-
learning resources exist, often free of charge. Two
of the leaders in the field of MOOCs are Udacity
Udacity was perhaps the very first MOOC and
was born out of a Stanford University experiment
by famous AI researchers Sebastian Thrun and Peter
Norvig (incidentally, Norvig is also co-author of the
AIMA course textbook (Russell and Norvig, 2010)).
They offered an online course called Introduction to
Artificial Intelligence online for free and managed
to achieve a simultaneous enrolment of more than
160,000 students in more than 190 countries.
EdX was founded by the Massachusetts Institute
of Technology (MIT) and Harvard University in May
2012 and provides courses from more than 60 univer-
sities, corporations and institutions.
We encourage students to enroll for free in one
or more of the courses provided by Udacity, and de-
mand that they enroll, also for free, in the edX course
CS188.1x Artificial Intelligence prepared by staff at
the University of California Berkeley. The courses
that we suggest build heavily on the textbook by Rus-
sell and Norvig (2010) and other textbooks on AI and
contain numerous interactive video lessons, self-tests
such as quizzes and multiple-choice tests, program-
ming exercises, and other material that encourage ac-
tive learning and are useful for TAI.
2.6 Workshops
All in-class learning activities in TAI take place in an
ordinary flat classroom on a single weekday that typ-
ically begins at 8:15 and ends at 16:00. However, the
teacher is neither required nor expected to be present
during the entire day.
The purpose of calling this teaching day a work-
shop is to emphasise an active, collaborative, student-
centred learning environment in which students will
construct their own learning, in contrast with more
teacher-centred learning environments. I will discuss
more about learning activities in the following sec-
3.1 Information and Structure
There are three important bits of information that
must be distributed to students before the very first
First, because the course is closely aligned with
the course textbooks, it is vital to let the students
know which books to buy as long as possible before
CSEDU 2017 - 9th International Conference on Computer Supported Education
the start of the semester. Unfortunately, some students
enrol in courses quite late, and we therefore have to
re-send this information several times to make sure
latecomers get it.
Second, we post a detailed 10-page course
overview, which serves as a reference guide for the
entire semester. This document includes core infor-
mation, high-level ILOs, teaching methodology, read-
ing material and topics, a weekly schedule, and infor-
mation about assignments, online resources, and the
course evaluation.
Third, I post the description of my first workshop,
with details on prerequisite homework (getting text-
books and registering on online resources) as well as
in-class learning activities. Unsurprisingly, many stu-
dents fail to do the homework and the workload of
the workshop is therefore intentionally smaller than
the remaining workshops to allow for this, with no
homework reading, exercises, or video lectures.
Early in the semester we also post a detailed list of
potential oral exam questions. As we elaborate on in
an accompanying paper submitted to this very confer-
ence (Osen and Bye, 2017), posting such a list serves
several purposes. It reduces uncertainty and stress re-
lated to the exam but just as important, it provides
students with mental hooks that aid in constructing
knowledge as they progress through the course, since
having read the questions will reduce the cognitive
load when new material is introduced, and they will
be able to more quickly putting new knowledge into
context and storing it in long-term memory.
All of the above information is posted on Fronter
in a highly structured way for easy and quick access,
even when using smartphones and tablets. The course
overview is rich in detail but is a document that we en-
courage students to refer back to regularly. It is useful
for the students to have a schedule of when all the dif-
ferent topics will be introduced as well as assignment
deadlines as early as possible so that students can plan
their studies and reserve time for assignments. Intro-
ducing the teaching methodology with the emphasis
on flipped classroom and active learning, with over-
arching ILOs and a well-defined main objective of
the course, also provides a useful top-down view that
guides the students during the course.
3.2 Workshops
Some time between a workshop and the next (typi-
cally 3–5 days), a new workshop description is re-
leased on Fronter. Each workshop description con-
tains details such as the date, classroom, start time,
teacher, and textbook chapters/topics to be studied;
the ILOs of that particular workshop; a rough sched-
ule of tasks to be completed; details on each task; and
homework for the next workshop. This homework
consists of reading selected chapters and solving se-
lected exercises in the textbook, as well as complet-
ing various tasks in the online edX course (see Sec-
tion 3.3) or on the code development platforms (see
Section 3.4).
Every workshop begins with a short evaluation of
the current status regarding problems students have
faced in their homework, questions they may have,
what they think about the teaching methods, and other
things that come to mind. Typically, there will be
some parts from the homework that students have
struggled with and have questions about. Minor ques-
tions are answered immediately, however, if it is
clear that a more thorough explanation is needed, the
teacher writes down the questions or topics on a list
and typically gives a micro-lecture on the various top-
ics afterwards.
Next, students are given a number of active learn-
ing tasks, varying in size. For example, a two-minute
task could be to discuss with another person in class
the possible answers of a question I give them, whilst
students can be working for up to several hours on an
larger exercises or assignments.
The teacher is not required to be available at all
times. Indeed, it may be beneficial to leave the class-
room at times, as internal discussion and group work
seem to flourish with the teacher absent.
3.2.1 Self-tests
For self-tests (quizzes or multiple-choice questions), I
usually ask everyone to do them individually, but en-
courage discussions with other students and myself.
This has the advantage that those who prefer some
time on their own to reflect on the answer get that,
whereas those who prefer to interact with others in
search of answers can do so. Afterwards, I ask stu-
dents in turn for answers, and most importantly, the
reasoning behind the answers. If the student’s reason-
ing is weak, I ask for clarifications from the rest class,
and finally, I often provide my own explanation, in
different wording.
Digital self-tests has the advantage that immediate
feedback is returned to the students. However, they
also require questions to be unambiguously phrased.
During a semester, there will very often be students
who are dissatisfied with a question, claiming that an-
other answer is more correct, or that there is no right
answer. This should not be viewed as something un-
fortunate. Rather, it is a golden opportunity for dis-
cussion in class and a driver for deep learning. Stu-
dents must have a thorough understanding of the topic
of the question and be able to formulate to the teacher
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence
why they disagree with the question, and other stu-
dents have to opportunity to delve into the discussion.
3.2.2 Oral Presentations
For oral presentations, students will generally be
working in groups, which tend to lower the fear of
presenting and generally improves the quality of both
quiz answers and presentations. Each group will ran-
domly be assigned a topic for the presentation, for ex-
ample, a particular search algorithm. Knowing that
they will have to make a presentation in class but not
knowing the topic beforehand likely acts as a cata-
lyst for completing the homework in a serious man-
ner. Also, making the presentation ad hoc in class and
explaining the topic to others force the students to or-
ganise their knowledge and insight on the topics.
Naturally, the students’ presentations will suffer
from the short preparation time in the classroom. I
therefore interrupt the presentations when needed and
first ask the presenting students if they are able to clar-
ify the issue at hand, and if not, I ask the class as a
whole if someone can help out. Even if I get a good
answer, I usually rephrase the answer with my own
words and try to be as concise and precise as possi-
ble, and hopefully, both the presenters and the rest of
the class get a much clearer picture of the particular
3.2.3 Textbook Exercises
Another activity during workshops include discussing
answers to textbook exercises. These exercises differ
from typical online exercises and self-tests, as they of-
ten require more work and are more time-consuming;
one often have to make an analysis of the problem
first, and then provide a long and elaborate answer. It
would therefore be a waste of time to do such exer-
cises in class, and instead, students are given a short
list of recommended textbook exercises to complete
at home before each workshop. In class, answers
to the exercises are walked through and discussed as
3.2.4 Coding Challenges
Finally, being a course on topics in AI, writing com-
puter code and implementing intelligent algorithms is
an inherent and important part of the course. Dur-
ing workshops, on average, a large portion of the time
is devoted to larger coding challenges contained in
the assignments, or minor coding challenges that are
solvable in a short period of time in class. More de-
tails on coding are provided in Section 3.4.
3.2.5 Competitions
To add some spice to the learning environment, I
sometimes arrange for small informal competitions in
the class, where the winner, or winning team, gets
nothing more than pride and glory. Sometimes I split
the class in groups and let each group work on the
same problem for 10-15 mins, say. The first team
who submits a working solution gets a point. Points
are then accumulated for various learning tasks during
the day. The students tend to like this concept but it
should not be overdone and detract from the learning
3.2.6 Discussions
Discussions are not a scheduled task in the workshop
but rather something that happens ad hoc during the
tasks mentioned above. Forcing students to discuss a
topic is rarely very awarding. Instead, having out-
sourced the traditional lectures from the classroom
provides more time to allow discussions to emerge as
needed. This is extremely useful for deep learning,
as it helps students learning by reducing the cognitive
load, delving deeper into topics, and allowing time to
organise and store new material in long-term mem-
ory. In addition, discussions will often sidetrack into
related topics, evident of students looking at the world
from a different perspective, discovering relationships
across topics, and even changing as a person.
3.3 Online AI Course
All students in TAI are required to enrol in an on-
line AI course offered by edX called CS188.1x Artifi-
cial Intelligence. This course has been archived since
2014, which means that no teaching staff maintain
or are active in the course. Nevertheless, all the re-
sources provided are available, free of charge, to edX
users. In Part A of TAI, four of the five topics are
covered by the edX course, with excellent video lec-
tures, step-by-step tutorials, and quizzes. The video
lectures come both in long and unedited versions and
in short and condensed edited versions that are also
interspersed with self-tests with immediate feedback,
all easily accessible on portable devices such as mo-
biles and tablets. The voices in the videos are tran-
scribed to text, which makes it easy for students to
quickly browse through a video by scrolling through
the text until a given point of interest.
Whereas some students favour watching the long
videos in one go and then use the the short, edited
videos for repetition, others jump straight to the edited
videos, especially if they have studied the textbook
first. The video format and interactive material also
CSEDU 2017 - 9th International Conference on Computer Supported Education
make it easy for students to squeeze in short micro-
sessions of homework, e.g., during a 15-min bus ride.
3.4 Coding Challenges
To construct knowledge, skills, and competence
within the AI topics that we study, students need to
practice on modelling problems with relevance to the
real-world and be able to select or modify existing in-
telligent algorithms from the AI literature or devise
new algorithms suitable to the level of abstraction in
their models. The next step is to implement the algo-
rithms and test them in a simulated environment, and
perform an analysis of their behaviour, often requiring
visualisations to understand what is going on.
Doing this from scratch requires both skills and a
lot of time, and henceforth we have outsourced this
to two code development platforms, HackerRank and
CodinGame, both of which provide excellent frame-
works within the students do not have to worry all the
necessary boilerplate code and instead focus on the
3.4.1 Intelligent Agents
Both platforms have been designed in a similar man-
ner where the user must input a computer programme,
commonly referred to as the AI, or bot, that reads data
from standard input, does some processing, and then
outputs a desired action to standard output. This loop
repeats for a number of time steps until the simula-
tion, or game, ends. Thus, the internals of the sim-
ulated environment behaves much like a black box
to the bot, however, certain parameters in the envi-
ronment are “sensed” and provided to the bot at each
time step, and in addition, each coding challenge pro-
vides details of the model the simulated environment
is based upon sufficient that a useful bot can be con-
structed. Hence, the coding challenges adopts the
same paradigm as the course textbook, in which AI
is centred around intelligent agents that sense, plan,
and act (Russell and Norvig, 2010).
Coding challenges typically revolve around cer-
tain themes in computer science and AI such as
queues, lists, search algorithms, path planning, con-
trol, etc. Very often, each challenge comes in several
versions, Easy, Medium, and Hard, in which the chal-
lenge becomes gradually more complex and difficult
to solve. An example is the Mars Lander challenge
on CodinGame, where students doing the Easy level
must design an intelligent agent, first for controlling
the landing of a spacecraft in the vertical dimension
only, then, for the Medium and Hard levels, in two
dimensions, which is exceedingly harder.
3.4.2 Design Process
When working on coding challenges, I emphasise
the importance of thinking about the problem before
jumping straight into coding. Together with the stu-
dents, we brainstorm various potential models of the
problem to be solved, and discuss aspects such as
levels of abstraction, and their advantages and lim-
itations. The solution method to be used is depen-
dent on the model, as an intelligent algorithm requires
the problem to be represented by exactly those com-
ponents that the algorithm was designed for. The
students then run a number of tests on the coding
platform to verify their design and receive automatic
feedback on which tests were passed or failed.
In contrast with HackerRank, the CodinGame
platform also provides visualisations of the simulated
environment. This is very valuable to students in the
iterative process of testing, debugging, refining, and
analysing their algorithms, because visual feedback
quickly can reveal problems in the execution of the
To conform with the requirement in NTNU’s Quality
System for Education that a course evaluation must
take place every time a course is run, we adopt a
scheme where all students present during workshops
(typically 7–8 students) belong to a so-called refer-
ence group. At the beginning of a workshop, we begin
with a class discussion about course-specific issues
such as status, progression, problems, etc., as well
as general issues you about the master programme in
general. The teacher takes notes, which form the basis
of an end-of-semester course evaluation report. These
short meetings provide an important arena and oppor-
tunity for students to help the teacher adjust course
early before issues become too big to fix.
In the sections below, I highlight some of the eval-
uation findings from the TAI course that was run au-
tumn 2016.
4.1 Workload
Students thought the overall workload was appropri-
ate, although perhaps slightly bigger than in some
other courses in the master programme. Some stu-
dents thought the workload for the assignments was
For this reason, and other minor reasons, we have chosen
to use the CodinGame platform exclusively starting from
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence
high but thought this was justified by the assignments
being highly valuable. For example, one student said
that (. . . ) the assignments were big but good. Even
though this course had a high workload, in retrospect
I would say that this was necessary, and the main rea-
son I learnt so much.
4.2 Teaching Methods and Learning
Students expressed enthusiasm for the workshop for-
mat and the elements of flipped classroom and had
little to add apart from comments that when revis-
ing theory from homework, e.g., in a micro-lecture, it
would be slightly better for the teacher to have slides
than having to move back and forth in the online
video lecture. Students also appreciated the fact that
videos were available in both short and long versions,
and different students sometimes favoured one over
the other. Students also expressed that the self-tests
were appropriate and useful, however, some multiple-
choice questions could contain ambiguously phrased
questions and therefore one could argue which answer
was really the correct one.
Such input is very useful for both the teacher and
for fellow students, as it triggers class discussion and
deeper insight into the problem that is being studied.
This is also an example of deep learning, commonly
observed with skilled students, as more shallow sur-
face learners will fail to note such ambiguities in the
Regarding the code development platforms, there
was consensus that HackerRank should not be used,
mainly because it was more difficult and lacks vi-
sualisations, in contrast with CodinGame. Students
also thoroughly enjoyed the gamification part of
CodinGame, where one achieves rating points and
badges and therefore are continuously eager to im-
4.3 Assignments
The students thought the topics covered in assign-
ments and the design of the problems to be solved
were appropriate, although the workload was quite
high. They also appreciated that assignments are first
released as draft versions and then modified by the
teacher and the students together through constructive
class discussions.
4.4 Structure and Information
The students expressed satisfaction with the course
structure and the continuous flow of information by
means of Fronter. The teachers have tried to encour-
age the Fronter forum functionality by posting an-
swers to several questions that students ask in class
or by email but there has been zero activity from stu-
dents in the forum. Students emphasised that being
on-campus and seeing each other regularly face-to-
face, there was not much need for an online forum.
Still, the teachers think the forum functionality should
be used but then acting more as a knowledge bank of
frequently asked questions.
4.5 Summary of Feedback
Throughout the course and in the weekly evaluation
sessions involving the entire class, students had very
little negative concern about the course itself, be it
teaching methods, workload, or curriculum. When
they had concerns, this was mainly related to clashes
of assignment deadlines across courses and sugges-
tions to reduce workload or modify content of assign-
Interestingly, students were very eager to discuss
issues of other courses and the master programme as
a whole. We see this as a sign that this course pro-
vides a safe and relaxing learning environment, since
students felt comfortable and very keen to share these
details with the teachers.
Finally, almost all students told us that they were
highly satisfied with the course, with two students ex-
pressing in writing that I also wish to express that
this has been one of the best courses I have taken in
higher education and [t]hank you for teaching one
of the better courses if not the best course I’ve had in
this master[’]s program[me].
In this paper I have presented a flipped classroom ap-
proach for teaching a master’s course on AI. Whilst
it should be clear that the course has been a great
success, the reader should by no means believe that
we have found the holy grail for teaching higher ed-
ucation courses. Indeed, there are several limiting
frame factors and challenges that must be overcome
for flipped classroom to work as intended. Below I
discuss some of these.
5.1 Frame Factors
The number of students in a class can be a limiting
frame factor for achieving ILOs. In the course pre-
sented here, run in 2015 and 2016, both cohorts con-
sisted of about 10 students, of which 2–3 quit early
CSEDU 2017 - 9th International Conference on Computer Supported Education
or never turned up.
If the class had exceeded about
25 students (which is also the number of seats funded
by the Ministry of Education), it may have been more
difficult to achieve the same fruitful learning environ-
ment in the classroom. For larger classes, one would
likely have to abandon flat classrooms and use large
auditoriums. University management would typically
see no problem in this and tell teachers to go ahead
and do their traditional one-way lectures as usual,
which is not desirable. Optionally, one could split
such large cohorts into smaller groups, which would
multiply the number of teachers required and asso-
ciated costs. This begs the question why there is a
transition in teaching methodology from classroom-
style teaching (one teacher per class of 30 students,
say) in lower education to auditorium-style lecture-
based teaching in higher education, thus turning stu-
dents into passive learners. Flipped classroom can
still be useful for large classes, however, but may
require much more effort in facilitating the in-class
learning activities.
Another important frame factor is the available
online resources. In TAI, we have been lucky to find
both an edX course that is closely aligned with the
topics in the course textbook for Part A, as well as
the CodinGame platform that provides a high quality
framework with a vast variety of AI coding challenges
suitable for achieving ILOs. Producing these online
resources oneself with the same quality, as well as
the material contained in textbooks, is an impossible
task for the common teacher. Still, with e-learning
as a buzz word nowadays, university management are
very eager for teaching staff to make videos to put on-
line. I think this approach is flawed, as it is extremely
time-consuming to produce video lectures oneself of
the desired quality, not to mention developing a suit-
able simulation environment for testing algorithms.
Instead, teachers should act as facilitators and devote
their time to finding such resources and tying them to-
gether in a didactically sound manner. For example,
the teacher must take care in selecting the right coding
challenges and guiding the students during problem-
solving (see the next section).
5.2 Problem-based Versus
Problem-solving Learning
Problem-based learning is a hot topic of higher educa-
tion and the interested reader may refer to our accom-
panying paper for reflections on many of its aspects
(Osen and Bye, 2017). Here, I refrain myself to a very
important finding by Hattie and Goveia (2013), who
15 students are enrolled in 2017.
upon examining 800 meta-analyses notes that
problem-based learning cannot be shown to have a
positive effect on achieving ILOs! The reason for
this, according to Sotto (2007), is the dinstinction be-
tween problem-based and problem-solving learning.
He argues that one should use a teaching approach
that employs well-designed case studies and avoid
problem-based and student-centred learning, espe-
cially for larger problems and when there is no clear
guidance towards how to solve them. Otherwise, stu-
dents will get stuck or spend too much time on finding
the necessary pieces to solve the puzzle by searching
the Internet or studying textbooks.
Adopting this distinction of Sotto (2007), I believe
the focus on problem-solving learning is a major key
to the success we have experienced in our course. For
example, the coding challenges on CodinGame are
carefully selected and much attention is given during
workshops on how to model the problems and how to
select appropriate solution methods.
On the other hand, if problems are too rigidly de-
fined, with perhaps only one solution approach pos-
sible, as when instructing students to follow detailed
step-by-step instructions, there is a danger that cre-
ativity is neglected and students only achieve surface
learning (Andersen, 2010). Comparing with bache-
lor’s courses, where such rigid schemes to some ex-
tent can even be desirable, master’s courses should
have a higher emphasis on being research-based and
investigative in nature.
In TAI, we seem to have found the right balance
between coding challenges being too rigid versus too
open-ended. There are nearly always several paths for
solving a given problem, and the teacher often pro-
vides a simple outline of a solution as a starting point
and then guides students towards improving the solu-
5.3 C-4 Dynamite for Learning
Central to my teaching approach is an emphasis on
four axes of learning, C-4, that together are dynamite
for learning: creativity, cooperation, competetition,
and challenge.
First, the learning activities we offer in the course
should facilitate creativity. In a creative process, stu-
dents model and design solutions to a wide variety of
problems but are usually not restricted (much) on how
to model the problem and which method to use.
Second, we employ many cooperative learning
activities in class and we encourage students to co-
operate with each other also for individual work.
C-4 is also a common plastic explosive.
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence
Third, as mentioned previously in Section 3.2.5,
I sometimes run small, informal competitions in
class as a driver for motivation. In addition, the
CodinGame platform is inherently competitive, as
users get rating points as they improve their solutions
or solve more problems. Students therefore compete
not only against others but also against themselves in
a quest for better ratings.
Fourth, we strive for students to critically chal-
lenge all the concepts they are introduced to in the
course and have indeed found that students tend to do
this more as the course progresses. Triggers for the
students can be as minor as mistakes in the written
formulations in quizzes, as discussed in Section 3.2.1,
or discovering that a solution method has certain lim-
itations that they first found out about when imple-
menting it on CodinGame. For example, a given intel-
ligent algorithm may not be able to provide a good an-
swer in reasonable time and a particular CodinGame
test therefore fails.
5.4 Future Work
In future versions of TAI, my colleague and I will look
both at the curriculum and on the learning activities
that we offer to bridge the gap between Part A and B
of the course. Most of what I have presented in this
paper relates to my own flipped classroom approach
for teaching of Part A, whereas my colleague has not
been able to find the same highly aligned and rele-
vant online resources to be able to facilitate flipped
classroom to the same degree for Part B. We will also
address the issue of colliding assignment deadlines
across parallel courses. Finally, we will be showcas-
ing this course to our colleagues as a motivator for
implementing flipped classroom themselves.
5.5 Conclusions
I have shown one way out of many that flipped class-
room can be implemented in a master’s course on AI.
The key to success is to be able to facilitate learn-
ing by cherrypicking textbook chapters and relevant
literature as well as suitable online resources such
as video lectures, self-tests, and coding challenges.
When such resources do not exist or lack in quality
or theory, one can try to make one’s own material,
e.g., a compendium supplementary to the textbook,
or one’s own video lectures. However, this is a very
time-consuming process if one wants to achieve the
desired quality, and should be a last resort.
I favour a teaching approach that has much in
common with constructive alignment, but emphasise
that the right balance between too rigidly defined lear-
ning activities where students are being “spoonfed”
and too vaguely defined problem-based learning ac-
tivities must be found, so that the problems are in-
vestigative in nature with different possible paths to-
wards solutions, yet at least one path must be reason-
ably easy for the students to find and follow.
The Software and Intelligent Control (SoftICE) Labo-
ratory is grateful for the financial support given by the
Study Committee at NTNU in Ålesund through the
project Research-based and Innovation-driven Learn-
ing (FILA), grant no. 70440500.
Abeysekera, L. and Dawson, P. (2015). Motivation and cog-
nitive load in the flipped classroom: definition, ratio-
nale and a call for research. Higher Education Re-
search & Development, 34(1):1–14.
Abu-Mostafa, Y. S., Magdon-Ismail, M., and Lin, H.-T.
(2012). Learning From Data: A Short Course. AML-
Andersen, H. L. (2010). “Constructive alignment” og
risikoen for en forsimplende universitetspædagogik.
Dansk Universitetspædagogisk Tidsskrift, 5(9).
Anderson, J. R. (2015). Cognitive Psychology and Its Im-
plications. Worth Publishers, 8th edition.
Biggs, J. (1996). Enhancing teaching through constructive
alignment. Higher Education, 32:347–364.
Biggs, J. (2011). Aligning teaching for constructing learn-
ing. The Higher Education Academy, York, United
Kingdom. Accessed online on 27.09.2011 at http://
Biggs, J. and Tang, C. (2011). Teaching for Quality Learn-
ing at University. McGraw Hill/Open University
Press, 4th edition.
Bishop, J. L. and Verleger, M. A. (2013). The flipped class-
room: A survey of the research. In ASEE National
Conference Proceedings, Atlanta, GA, volume 30.
Bligh, D. A. (1998). What’s the Use of Lectures? Intellect
Bonwell, C. C. and Eison, J. A. (1991). Active Learning:
Creating Excitement in the Classroom. 1991 ASHE-
ERIC Higher Education Reports. ERIC.
Borkowski, J. and Thorpe, P. (1994). Self-regulation and
motivation: A life-span perspective on under- achieve-
ment. In Schunk, D. and Zimmermann, B., editors,
Self-regulation of learning and performance: Issues
of educational applications, pages 44–73. Hillsdale,
NJ: Erlbaum.
CSEDU 2017 - 9th International Conference on Computer Supported Education
Bowen, C. W. (2000). A Quantitative Literature Review
of Cooperative Learning Effects on High School and
College Chemistry Achievement. Journal of Chemi-
cal Education, 77(1):116.
Clark, R. C., Nguyen, F., and Sweller, J. (2011). Efficiency
in learning: Evidence-based guidelines to manage
cognitive load. John Wiley & Sons.
Entwistle, N. and Ramsden, P. (1983). Understanding stu-
dent learning. Beckenham: Croom Helm.
Felder, R. M. and Brent, R. (2005). Understanding stu-
dent differences. Journal of Engineering Education,
Foldnes, N. (2016). The flipped classroom and cooperative
learning: Evidence from a randomised experiment.
Active Learning in Higher Education, 17(1):39–49.
Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K.,
Okoroafor, N., Jordt, H., and Wenderoth, M. P. (2014).
Active learning increases student performance in sci-
ence, engineering, and mathematics. Proceedings
of the National Academy of Sciences, 111(23):8410–
Gynnild, V. (2001). Læringsorientert eller ek-
samensfokusert? Nærstudier av pedagogisk
utviklingsarbeid i sivilingeniørstudiet. PhD the-
sis, NTNU.
Gynnild, V., Holstad, A., and Myrhaug, D. (2007). Teach-
ing as coaching: A case study of awareness and learn-
ing in engineering education. International Journal of
Science Education, 29(1):1–17.
Gynnild, V., Holstad, A., and Myrhaug, D. (2008). Identi-
fying and promoting self-regulated learning in higher
education: roles and responsibilities of student tu-
tors. Mentoring & Tutoring: Partnership in Learning,
Hattie, J. and Goveia, I. C. (2013). Synlig læring:
et sammendrag av mer enn 800 metaanalyser av
skoleprestasjoner. Cappelen Damm akademisk.
Johnson, D., R., J., and Smith, K. (1998). Active Learn-
ing: Cooperation in the College Classroom. Interac-
tion Book Co., Edina, MN, 2nd ed. edition.
Lage, M. J., Platt, G. J., and Treglia, M. (2000). Invert-
ing the classroom: A gateway to creating an inclusive
learning environment. The Journal of Economic Edu-
cation, 31(1):30–43.
Lan, W. (1996). The effects of self-monitoring on stu-
dents’ course performance, use of learning strate-
gies, attitude, self-judgment ability, and knowledge
representation. Journal of Experimental Education,
Marshall, D., Summers, M., and Woolnough, B. (1999).
Students’ conceptions of learning in an engineering
context. Higher Education, 38(3):291–309.
Marton, F. (1981). Phenomenography Describing con-
ceptions of the world around us. Instructional Science,
Marton, F. and Booth, S. (1997). Learning and Awareness.
Mahwaw, NJ: Lawrence Erlbaum.
Negnevitsky, M. (2005). Artificial Intelligence: A Guide to
Intelligent Systems. Addison Wesley, 2nd edition.
Osen, O. L. and Bye, R. T. (2017). Reflections on Teach-
ing Electrical and Computer Engineering Courses at
the Bachelor Level. In Proceedings of the 9th Inter-
national Conference on Computer Supported Educa-
accepted for publication.
Piaget, J. (1968). Six psychological studies. Trans. A. Ten-
Prince, M. J. (2004). Does active learning work? A review
of the research. Journal of Engineering Education,
Prosser, M. and Trigwell, K. (1999). Understanding learn-
ing and teaching: The experience in higher education.
Buckingham: Society for Research in Higher Educa-
tion and/Open University Press.
Russell, S. and Norvig, P. (2010). Artificial Intelligence:
A Modern Approach. Pearson: Upper Saddle River,
New Jersey, 3rd (international) edition.
Russell, S. and Norvig, P. (2013). Artificial Intelligence:
A Modern Approach: Pearson New International Edi-
tion. Pearson: Upper Saddle River, New Jersey, 3rd
Schaathun, H. G. and Schaathun, W. A. (2016). Learning
mathematics through classroom interaction. In The
18th SEFI Mathematics Working Group seminar on
Mathematics in Engineering Education, pages 155–
Schroeder, C., Scott, T. P., Tolson, H., Huang, T.-Y., and
Lee, Y.-H. (2007). A Meta-Analysis of National
Research: Effects of Teaching Strategies on Student
Achievement in Science in the United States. Journal
of Research in Science Teaching, 44(10):1436–1460.
Sotto, E. (2007). When teaching becomes learning: A the-
ory and practice of teaching. Bloomsbury Publishing.
Springer, L., Stanne, M., and Donovan, S. (1999). Effects of
Small- Group Learning on Undergraduates in Science,
Mathematics, Engineer- ing and Technology: A Meta-
Analysis. Review of Educational Research, 69(1):21–
Topping, K. and Ehly, S. (1998). Peer-assisted learning.
Topping, K. J. (1996). The effectiveness of peer tutoring in
further and higher education: A typology and review
of the literature. Higher education, 32(3):321–345.
Vygotsky, L. (1978). Mind in society. London: Harvard
University Press.
The Teacher as a Facilitator for Learning - Flipped Classroom in a Master’s Course on Artificial Intelligence