Sunday, December 30, 2018

E-Learning, Digital Learning, Digitally Transformed Learning---Do They All Mean the Same?

Nowadays we often hear that the term "e-learning" is a bit outdated, we should rather speak about "digital learning" or even "digitally transformed learning" when we talk about things like using learning management systems (LMS), students collaborating online, etc. Are these terms interchangeable, then? Does it even mean that in learning we are ahead of other areas which are yet to be digitally transformed---right now or in the near future---as we have been doing e-learning for 15 or 20 years now, or perhaps even since the 1960s (have look at PLATO)?

As for other fields, you could talk about the aspects of digitization, digitalization, and digital transformation of learning, and what they imply when it comes to skills and competences you should have and you could acquire. We will do this in a separate article.

In an attempt to look at similarities with other areas, we could try to define various waves or eras of learning, or even try to define "learning revolutions" in analogy to "industrial revolutions." And then we would arrive at terms like "Learning 4.0" (to have the same version number as in "Industry 4.0," or rather "Learning 3.0," or maybe "Learning 3.11 for Workgroups---OK, just kidding). And then we would need cover terms to name the eras of learning.

When we take the industrial revolutions as tertium comparationis, we have the first revolution with the advent of looms---mechanical work done by hand had been automated on a small scale---and the steam engines---the automation of mechanical work at a larger scale. The second revolution was the advent of electricity. This also involved the introduction of the production of electricity as a utility, as a service. It became possible to produce energy at some place and transfer it over fairly long distances to run machines. It wasn't necessary any longer to produce energy directly in or very close to factories. Ford introduced assembly lines and mechanical work done by hand changed again, workers specialized in specific areas. The third revolution came with the introduction of computers, we got CAD/CAM, industrial robots, etc. In all those processes, the human was the main factor: humans control and regulate, they make decisions with the help of machines.

Now in the fourth revolution we face the merging of real and virtual worlds, we not only interact with machines, we let machines decide and call this digital transformation. Computer programs decide whether or not you are creditworthy---some years ago, the banker would inspect the gathered and aggregated data and then make a decision; now the computer decides on its own based on models it created from relevant and irrelevant data using machine learning. We are close to let machines decide whether or not you are prone to return to your bad habits after rehab or prison (see, e.g., AI Judges and Juries in the December issue of CACM).

Let's look at education, where can we position "e-learning"? With the advent of tele-learning in the 1950s and 1960s, we find some aspects of automating parts of teaching. But even earlier, in the 1920s, we have actual machines: mechanical devices as first introduced by Sidney Pressey to let people answer multiple choice questions. Later Skinner developed them further to provide automatic, immediate, and regular reinforcement, and thus trigger learning. And it could be shown that students actually learned while using those machines. Already then we find the discussion whether or not machines would replace human teachers in the near future.

When we look at developments in the 1970s, with the PLATO systems, we find the same ideas: to provide automatic, immediate, and consistent feedback. That's part of "teaching," though, it doesn't redefine "learning"!

Then, at the start of the 21st century, we integrated computers into teaching and learning. We often talk about "e-learning," but we only rarely talk about "e-teaching." However, even with LMS and all of their still improving (or let's rather say: accumulating) functionalities, we still focus on automatic, immediate, and consistent feedback. That's what all the e-assessment, peer activities, forums, etc. are about. And we all agree that just using your fully-fledged LMS to distribute your PowerPoint slides doesn't qualify to be named "e-learning." But still, we have no actual interaction of human and machine, you just get feedback and then decide what to do next. So "e-learning" in this sense is just a contemporary (as in "use mobile electronic devices") teaching machine.

But wait, we also had intelligent (adaptive) tutoring systems in these first years of e-learning! Actually, those teaching machines by Skinner as well as the PLATO V were also intelligent tutoring systems (ITS)---and they were advertised as such. So also this isn't a brand new idea! For various reasons, these systems weren't successful at the time. But most contemporary e-learning research doesn't refer to those old publications when talking about adaptive systems. But maybe now, given the available computing power, it would be time to revisit those old ideas. If technology (including bandwidth) for distribution and interaction was the bottleneck back then, we may be able to solve those issues now.

There have been various attempts around the start of this century, though, tackling another potential bottleneck: the learner model. Using computers, it was more comfortable to implement and maintain various learner models accounting for different learning paths through the material towards the final goal of acquiring some specific competencies or skills. Attempts like <ML>³, (Multidimensional Learning Objects and Modular Lectures Markup Language) or elml (eLesson Markup Language) aimed at foreseeing learning paths and provide students with the appropriate next steps depending on previous actions and (formative) testing outcomes. It turned out that creating such material was rather challenging and demanding. The same was true for testing formats like SET (Satzergänzungstests), which allow you to answer a question by adding parts of sentence(s), an instantiation of "Reihenerweiterungswahl" (Closed Sequence Selected Extension Items) according to the typology of Rütter (Rütter, T., 1973. Formen der Testaufgabe. Eine Einführung für didaktische Zwecke. C. H. Beck, München.) as we showed in a paper. Even with a rather sophisticated editor, it was a nightmare to produce those tests.

Writing learner models using rules, manually, is probably not working. One simply cannot foresee all possible activities and interests. A truly adaptive intelligent system would need a model covering all those possibilities. For now this seems achievable only by using machine learning. In the way we construct language models we could create learner models to feed into tutoring systems and let the machine decide what experience the learner makes next, what problems the learner should solve next, etc. And there we would have it: the digitally transformed teaching/learning as a blending of real and virtual worlds with the machine not only providing information to support human decision-making but with the machine deciding and interacting with the human. Of course this also raises ethical questions: is it OK to have the computer model you as a learner? But that's along the same lines as in "is it OK to have the computer model you to decide whether or not you will get this credit or whether or not you can get that life insurance?"

As long as we identify "e-learning" with "using the full potential of LMS and (apps on) mobile devices" (does anybody remember "clickers"? You can have them as apps now, yeah!), we don't talk about the digital transformation, but about the electronic re-engineering of teaching machines. But as long as we're just deploying "electronic teaching machines," we should stick with the term e-learning. Oh, and we still have vast communities who use LMS as PDF or PowerPoint distribution vehicles only, there isn't even digitalization involved, only digitization.

Clearly, with all those MOOCs around where you interact with the video and the instructors/tutors, a lot of logging is could be going on. This data will be used to model learners. And as for language models in Natural Language Processing, those models created by machine learning might be not exact but appropriate or good enough for specific tasks. The big issues there revolve around the questions of "which features matter, which features do you use?" The same will be true for learner models or learning models. What we have in e-learning are various models of teaching, and those could be described by manually crafted rules. They are based on hundreds of years of research and developments in didactics and pedagogy (and schools thereof). For determining and weighting features for learning, we shouldn't leave the fields to the usual suspects of Big Data processing. This research and development and thus the digital transformation of learning should be driven by the field of teaching and learning, by the experts involved with didactics and pedagogy.

No comments: