Presentation at Online Learning 2018/I4PL in Toronto, October 2018

1. Using AI in Higher Education
Stephen Murgatroyd, PhD
Chief Innovation Officer,
Contact North | Contact Nord

4. This Presentation
Outline the
Potential of AI for
Education
1
Share Current
Developments
and Issues
2
Look at the Future
3

5. 10 Ways in Which AI Could
Impact Higher Education
“The future isn’t what it used to be…” Yogi Berra

6. 1. Natural Language
Generation
 Natural Language Generation: Producing text from
computer data. Currently used in customer service,
report generation, and summarizing business
intelligence insights. Sample vendors: Attivio,
Automated Insights, Cambridge Semantics, Digital
Reasoning, Lucidworks, Narrative Science, SAS,
Yseop.
 Imagine in Education: Automatically generating
assignment questions, course materials, teaching
resources.

7. 2. Speech Recognition
 Speech Recognition: Transcribe and transform
human speech into a format useful for
computer applications. Currently used in
interactive voice response systems and mobile
applications. Sample vendors: NICE, Nuance
Communications, OpenText, Verint Systems.
 Imagine in Education: Being able to support
learners who find following conversation
difficult because of hearing difficulties;
enabling the reconstruction of important
conversations between researchers.

8. 3. Virtual Agents
 Virtual Agents: From simple chatbots to advanced
systems that can network with humans. Currently used in
customer service and support and as a smart home
manager. Sample vendors: Amazon, Apple, Artificial
Solutions, Assist AI, Creative Virtual, Google, IBM, IPsoft,
Microsoft, Satisfi.
 Imagine in Education: Expanding the use of chat-bots to
link to adaptive learning engines so that students are
able to be engaged in adaptive learning. They are also
being used to provide 24/7 technical support and to
help students navigate the administrative requirements
of school, college and university programs.

9. 4. Machine Learning Platforms
 Providing algorithms, APIs, development and training toolkits, data, as
well as computing power to design, train, and deploy models into
applications, processes, and other machines. Currently used in a wide
range of enterprise applications, mostly involving prediction or
classification. Sample vendors: Amazon, Fractal Analytics, Google,
H2O.ai, Microsoft, SAS, Skytree, 24[7].
 Imagine in Education: A concrete example of machine learning in use
is McGraw-Hill Education’s ALEKS, a web-based, intelligent assessment
and learning system, which uses graph theory to break up a domain
into concepts. The edges of the network might then be used as a
bridge to another domain. For instance, if algebra has 500 concepts, a
student might be tested when he or she begins learning, pinpointing
exactly what a student knows and doesn’t know, and then creating an
appropriate path through the learning domain that is selected based
upon the start point and then continuously revising the learning map
for that individual student.

10. 5. AI Optimized Hardware
 AI-optimized Hardware: Graphics
processing units (GPU) and
appliances specifically designed and
architected to efficiently run AI-
oriented computational jobs.
Currently primarily making a
difference in deep learning
applications. Sample vendors:
Alluviate, Cray, Google, IBM, Intel,
Nvidia.
 Imagine in Education: Faster
simulations for science and
technology programs and more
accurate graphic and colour images
for arts.

11. 6. Decision Management
 Decision Management: Engines that insert rules and logic into AI
systems and used for initial setup/training and ongoing
maintenance and tuning. A mature technology, it is used in a
wide variety of enterprise applications, assisting in or performing
automated decision-making. Sample vendors: Advanced
Systems Concepts, Informatica, Maana, Pegasystems, UiPath.
 Imagine in Education: Enabling critical decision making about
the student journey – risk assessments (drop-out, non-
completion, program switching, remediation requirements),
providing instant supports “just in time”.

12. 7. Deep Learning Platforms
 Deep Learning Platforms: A special type of machine learning
consisting of artificial neural networks with multiple abstraction
layers. Currently primarily used in pattern recognition and
classification applications supported by very large data sets. Sample
vendors: Deep Instinct, Ersatz Labs, Fluid AI, MathWorks, Peltarion,
Saffron Technology, Sentient Technologies.
 Imagine in Education: Deep learning engines can generate learning
pathways and resources to support specific learner needs, once the
engine has mastered the learning outcomes expected of learners. A
student struggling with a specific set of issues within a subject can
have remediation from the AI system generated automatically so
that the student can “get back on track” and master the knowledge
and skills they need to progress.

13. 8. Biometrics
 Biometrics: Enable more natural interactions between humans and
machines, including but not limited to image and touch recognition,
speech, and body language. Currently used primarily in market research.
Sample vendors: 3VR, Affectiva, Agnitio, FaceFirst, Sensory, Synqera,
Tahzoo.
 Imagine in Education: Biometrics are already used as part of the remote
proctoring systems for student exams and assessments; fingerprint, facial
recognition and writing patterns. Fingerprints are used by some school
systems as a way of automating payments, library borrowing and access
to IT services.

14. 9. Robotic Process Automation
 Robotic Process Automation (RPA): Using scripts and other
methods to automate human action to support efficient
business processes. Currently used where it’s too expensive or
inefficient for humans to execute a task or a process. Sample
vendors: Advanced Systems Concepts, Automation Anywhere,
Blue Prism, UiPath, WorkFusion.
 Imagine in Education: The University of Colorado Boulder is
using robots fitted with telepresence screens to connect
students temporarily unable to attend classes (due to illness,
injury or some other reason) to their instructors and peers.
Telepresence robots are remotely controlled video
conferencing devices that are increasingly used in fields such
as business and healthcare. Robots are also being used to
help autistic students develop social and communication skills
(Cheung, et.al, 2018; Van Hooijdonk, 2017).

15. 10. Text Analytics
 Text Analytics and NLP: Natural language processing (NLP) uses and
supports text analytics by facilitating the understanding of sentence
structure and meaning, sentiment, and intent through statistical and
machine learning methods. Sample vendors: Basis Technology, Coveo,
Expert System, Indico, Knime, Lexalytics, Linguamatics, Mindbreeze,
Sinequa, Stratifyd, Synapsify.
 Imagine in Education: The application of NLP in an education system is
very effective for the analysis of errors in objective assessments and for
the assessment of essays and complex answers to questions. Various
linguistic approaches and tools can be utilized for analyzing the errors
such as grammatical and stylistic errors, errors in fact or problems of
understanding and competency. Teachers can easily mark these errors in
the papers of students, but so can AI systems with a reliability of 92%
when compared to human markers. Some 60,000 Chinese schools are
using such systems for grading and marking assignments (Alger, 2018).

16. Current AI Activities in Higher
Education
Complete Summary at www.teachonline.ca

17. 1. Automated
Feedback and
Grading
The Open
University UK’s
Open essayist
To support students in drafting essays, a team from the
Open University developed OpenEssayist, an intelligent
linguistic analytics tool used in real-time to analyze text
in essays and generate automated feedback.
OpenEssayist analyzes 3 aspects of an essay; structure,
key words/phases and key sentences and then presents a
summary to the student that conveys the key points of
the essay, allowing students to; 1) reflect on the draft text
2) review how the essay is organized 3) understand how
the key terms are being used across the essay and how
they combine to form a cohesive discussion. Evaluation
of OpenEssayist shows the number of drafts submitted to
the tool has a positive impact on the grades awarded for
the first assignment. The cohort of students using
OpenEssayist achieved significantly higher overall grades
than the students in the previous cohort.

18. 2. Intelligent
Tutoring @ The
University of
California, San
Diego
The University of California, San Diego, computer science
Professor Pavel Pevzner and colleagues designed the MOOC
course Introduction to Genomic Data Science as an adaptive
Intelligent Tutoring System (ITS) for the edX platform. To lead
students through a personal learning pathway, allowing
evaluation at every stage of the student’s learning, the course
has incorporated quizzes and “just in time” exercises to assess if
the student has understood the content. If a student answers
incorrectly, they are directed to a remedial site to assist with
content understanding and attainment. The course also
provides coding challenges to assess each student’s progress
and to replace the basic multiple-choice quizzes prevalent in
most MOOCs. Integrating an ITS into a MOOC format
addresses the big issue of course modification and updating, as
the ITS platform is designed to support easy, continual
updating, especially useful in topics where students experience
difficulty.

19. 3. Learning
Analytics
@ The Arab
University
(Kuwait)
One of Arab Open University’s (AOU) main
objectives is to double the number of students over
the next 5 years. To accomplish this, they are using
IBM Watson Analytics to improve student retention
and identify students at risk of dropping out. IBM
Watson Analytics uses artificial intelligence
and machine learning algorithms to sense, predict,
infer and provide recommendations. With insights
revealed by the analytics, AOU can identify
vulnerable students by pinpointing key drivers of
student attrition. One of the most promising
outcomes is the decision support dashboards, such
as the one related to Student Risk Factor (SRF), a
score which is composed of the student’s current
GPA, progression rate, and the number of warnings
received. By identifying at risk students, the
dashboard acts as an early alert system, enabling
the AOU management to take corrective actions
and targeted initiatives to help struggling students
get back on track for success — increasing
retention and boosting student numbers.

20. 4. Student
Support
Services @
Deakin
University
Cognitive-computing technology, available 24/7, allows students
at Deakin University to ask IBM Watson questions about
administrative and course information in natural language in
place of searching through keyword-based FAQs. Watson is
asked over 1600 questions a week about an assortment of
topics, such as admissions, enrollment, tuition and fees, financial
assistance, student housing, extracurricular skills development,
health and wellness, facilities, job placement, employment
preparation, job skills assessment and academic help. Deakin
University predicts the use of this tool will boost enrollment by
up to 10% as a result of a 20% increase in student satisfaction.

21. 5. Adaptive
Group
Formation @
Universitat
Politècnica de
València (Spain)
Higher education institutions are recognizing the pivotal role
they play in developing essential teamwork skills among
students to help prepare them for success in the workplace,
using skills such as positive team dynamics, clear
communication and interpersonal collaboration. The Universitat
Politècnica de València in Spain addressed this need by
developing a tool to create diverse, classroom-based teams
grounded in theory that identifies eight different behavioural
patterns and roles of successful teams. The tool calculates and
proposes optimal team configurations and gathers feedback
from team members on the roles of team members to be used
in future team formation tasks. The team formation tool is more
successful than traditional team methods in facilitating different
teamwork aspects, such as student satisfaction, team dynamics,
and cooperation and coordination

22. 6. Virtual
Agents @
Georgia Tech
A Georgia Tech computer science professor created a virtual
teaching assistant (TA) named Jill Watson, based on the IBM
Watson platform, to help answer the more than 10,000 forum
posts in his online course of over 300 students. The virtual TA
takes routine essential questions , such as queries about proper
file formats, data usage, and the schedule of office hours -
questions with firm, objective answers, while the human TAs
handle the more complex questions. For the first few weeks, Jill
Watson gave irrelevant answers which were posted to a site only
the developers could see. They worked to overcome Jill’s issues
and soon her answers had 97% certainty and they were sent
directly to students with no intervention or vetting by
developers. Students were unaware their TA was actually a
computer.

23. 7. Virtual Reality
@ Queens
University
(Canada)
Ensuring students acquire comprehensive expertise with
advanced theoretical knowledge and soft professional skills is a
top priority for professional schools in law, engineering,
medicine and business. This is why Queen’s University is
using Ametros Learning’s intelligent simulations powered by IBM
Watson’s cognitive-computing tool to focus on case-based
teaching through simulations of real-world challenges, allowing
students to develop and hone decision-making and problem-
solving abilities. Students use textual, visual and oral
communication on the platform to practice “real”
communication interactions with artificially intelligent characters.
These characters can take on the role of a client, vendor, patient,
peer, and/or team member. Characters involve students in
contextual interactions and provide individualized feedback on
student work. The simulation platform creates a rich, risk-free
simulated environment in their chosen field, where students
learn through experience

24. 8.Personalized
Adaptive
Learning
Environment @
University of
Zagreb
Spaced repetition is a learning technique that uses
increasing intervals of time between learning new content
and the review of previously learned content. It is usually
applied in language learning to accommodate the large
amount of content to be retained in the student’s long-term
memory. Software can adapt to the student’s prior
knowledge and use that information to help the student
memorize characters, vocabulary, and phrases by
determining how frequently the student must review content
so it stays in their long-term memory. Faculty at the
University of Zagreb harnessed this technique to teach
Japanese to a group of 27 undergraduate students through
the language learning application, Memrise. Students who
regularly used Memrise for learning Japanese had 40%
better grades than those who didn’t at the end of two
consecutive semesters.

25. 9. Online
Proctoring @
Stevens Institute
of Technology
 A team from Stevens Institute of Technology’s School of
Engineering and Science piloted a virtual laboratory tool
with biometric authentication used to identify the student
and monitor their actions via remote proctoring using facial
recognition techniques to an undergraduate mechanical
engineering course. When using the virtual laboratory tool,
the students log in by scanning their faces with a web
camera. While performing a laboratory assessment, the
student sits in front of the camera and the virtual laboratory
tool monitors their facial expressions and head motions in
order to identify suspicious behaviours. Upon detection of
such behaviours, the tool records a video for further analysis
by the laboratory administrator. The virtual lab tool with a
virtual proctor works well and provides a high degree of
accuracy in detecting suspicious behaviour during
assessments.

26. Some Issues
and Concerns

27. Key Concerns
 Our inability to interrogate some AI systems to determine just how decisions within the system
(outcomes) relate to inputs – what are the algorithms in use and how do they actually function?
We call this the “explain-ability” challenge
 Linked to this is the issue of bias – when data is being used to create algorithms, what kind of
bias is implicit in the system? For example, it is already the case that a great deal of our
understanding of coronary heart disease prior to 2000 was based largely (though not exclusively)
on studies of men. It was assumed that our understanding of these data would apply equally to
women and could be interpreted, with modest adjustments, for children and younger adults.
Subsequent work has suggested that our prior understanding was limited by this bias. Data
diversity is a key requirement for effective AI.

28.  Linked to this is the challenge of the “filter-bubble” – already an issue with social media. When
social media filters information delivered to the user by past choices and decisions (e.g. Amazon
book recommendations, Facebook tailoring feeds to suit patterns of choices made by the user), we
begin to filter out messages, ideas, resources that do not fit the profile of the user. Vegans do not
see meat recipes, fans of tango rarely hear ballet music.
 There is also an accountability-responsibility problem. When a piece of software fails – for
example, GPS software, it is possible to determine whether the problem is a hardware problem, a
software problem, a data-entry problem or a human error problem. With AI systems, this is more
difficult, since many of these sharp distinctions (software, data, human input) are blurred. When a
self-driving car kills a pedestrian, for example, who is to blame? The system for making a choice
between the driver’s safety and that of the pedestrian, the human in the vehicle, the pedestrian?
Some have also referred to this as the “artificial stupidity problem” (Bossman, 2016).

29. and
 Privacy is a key issue, especially given the huge appetite for data which AI systems appear to have. For example,
there is the case of a retailer who was able to “know” that a teenage girl was pregnant before her parents did (Hill,
2012) or if unintended data-sharing between different parts of the UK’s National Health Service (NHS), breeching
several policy and procedural guidelines (Hodson, 2016).
 Bad actors are also a major and growing problem for AI systems. Despite many with good intentions, not all uses of
AI are benevolent. There’s also the issue of bad actors, of both governmental and non-governmental nature, that
might put artificial intelligence and machine learning to ill use. A very effective Russian face recognition system was
deployed in 2017 and it proved to be a useful tool for an oppressive regime seeking to identify and crack-down on
dissidents and protestors. Such a system could also be used to “out” a gay person, especially dangerous in those
nations where being gay is a criminal offence. Another machine learning algorithm proved to be effective at peeking
behind masked images and blurred pictures. Other implementations of AI and ML are making it possible to
impersonate people by imitating their handwriting, voice, and conversation style, an unprecedented power that can
be more than useful in a number of dark, criminal scenarios.

30.  The interface challenge – or managing the singularity is a substantial challenge. As more and
more people live their lives in partnership with technology, how do we manage both the
people:technology interface and the people:people interface? We have already seen major
in social relationships (especially amongst iGen teens) – children now compete with technology
the parents’ attention – as AI systems become sophisticated and we depend more and more on
them, our relationship with each other and with technology is being challenged.

31. My Conclusion
One task we each
will have is to learn
to dance with
robots! We will all be
cobots!