Page 69 - ITU Journal, ICT Discoveries, Volume 3, No. 1, June 2020 Special issue: The future of video and immersive media
P. 69
ITU Journal: ICT Discoveries, Vol. 3(1), June 2020
on the basis of this information. This way, every Ideally, assessment should include continuous and
student can have their own personalized learning objective measurements of how students are
experience. progressing toward targeted goals without the
student being aware of the assessment. This has
The evaluation of intelligent tutoring systems like
AutoTutor has been positive. In a series of been aptly called stealth assessment [49].
experiments the learning gains of students There has been an increasing amount of evidence
interacting with the intelligent tutoring system that demonstrates the learning process can be
were on a par with those of human tutors, which measured non-intrusively, online, and objectively
were significantly better than those after students using sensing technologies [48, 50]. Sensing
studied a textbook on the same topic [43, 44]. technologies consist of sensors which are able to
measure neurophysiological, cognitive and affective
Yet, despite the promise of these systems, their processes. These sensors include, for example eye-
embodiment has generally been that of a cartoon tracking, recordings of language and speech,
character with very rudimentary facial expressions measures of head movements and measures of
and gestures and highly computerized speech heart rate or brain activity such as
synthesis, with written input from the student.
electroencephalography (EEG). They are thought to
Modern day techniques such as photogrammetry be able to detect changes in how information is
(which generates a 3D model from pictures) now processed and therefore in how cognitively
allow us to scan in individuals, in the case of demanding a task is, an important transition
SpaceBuzz the astronaut of the respective country, occurring during learning [51, 52]. Indeed, a recent
and use artificial intelligence algorithms to animate meta-analysis examining 113 experiments on non-
the face and gestures of the agent in a natural way invasive neurophysiological measures of learning
so that it becomes hardly distinguishable from demonstrated that these measures yield high effect
human non-verbal behavior [45]. Developments in sizes and are valuable for assessing learning.
speech recognition allow for speech input, so the Additionally, learning analytics and educational
conversation becomes more natural, and speech data mining has shown success in detecting affect
synthesis allows for recreating the voice of an actual related to learning gains [48, 53].
individual [46]. Computational linguistic As technological advancements take place, we
techniques in the meantime, allow for sophisticated expect the validity and applicability of assessment
syntactic parsing, distributional semantics and through sensing technologies to increase. For
dialog management [47]. example, solutions for (automatically) dealing with
The enhanced embodiment of the agent provides noise and advanced open software tools for aiding
lots of promise to these systems in the field of in analyses of these types of datasets are available
immersive education. AI techniques operate the and are being enhanced [54], even supporting real-
non-verbal and verbal behavior of the intelligent time analysis and monitoring [55]. As another
tutoring system. Machine learning mechanisms example, wearable, portable, and compact sensors
allow these systems to learn over time, both reduce set-up time and associated costs, and
improving their conversational skills as well as their increase comfort, making wearables very attractive
knowledge and pedagogy. to apply for gaining insight in learning.
Taking this one step further, embedding sensing
6. LEARNING ANALYTICS technologies in learning systems could progress
Current educational practice assesses students applicability even more. Such systems have the
periodically, either without corrective measures, as potential to assess learning states and provide
in Bloom’s conventional learning, or with corrective instantaneous assistance or adaptation of the
measures, as in Bloom’s mastery learning. There are learning material based on the recorded data [56].
a number of important downsides to this current As an example, Sood and Singh [57] have provided
approach to assessment [48]: 1) It might interrupt a framework on how to combine virtual reality and
and even negatively influence learning; 2) Little measures of brain activity in a serious game for
insight is provided into the learning process itself; robust and resilient learning and an enhanced
3) When assessment demonstrates suboptimal learning experience.
performance which would benefit from corrective
procedures, intervening may already be (too) late.
© International Telecommunication Union, 2020 47