e-Learning Ecologies MOOC’s Updates

Essential Update #7: The Promises and Limits of Adaptive Learning

Philip Kerr, in his 2014 short booklet, Adaptive Learning, which is a review of adaptive learning in English language teaching, hits on a couple of key critiques of adaptive learning.

 

Before we get there, however, it’s useful to define Adaptive Learning (AL): Kerr never truly defines it, instead offering examples of adaptive learning platforms in the marketplace (notably Duolingo and Cerego’s iKnow!). These systems are characterized by elements of gamification (levels, point systems) as they use technology to track student performance and serve up content based on responses. As a user of Duolingo, I know that an incorrect answer on a given problem will cause that problem to reappear until I (the student user) have correctly answered the problem in multiple formats (e.g., given an English sentence, choosing French words from a bank of available words; typing out the English translation of the French phrase; and given audio of the French phrase, typing the phrase in French). This useful video by OpenOsmosis.org uses a simple definition: adaptive learning occurs when the “material a student receives changes based on their individual performance…[and] is often automated and driven by technology” (OpenOsmosis.org).

 

Kerr rails against the overuse of “adaptive,” especially among providers of online learning materials or solutions; working for some of these providers, I’ve seen that the term “adaptive” has been used to describe almost any personalization or customization possible in a given learning platform--particularly in 2014 when this term was at the zenith of its trendiness. A robust adaptive system has hugely exciting potential--technology could help teachers truly personalize learning experiences for their students’ unique learning preferences and styles.

 

But the thing about an adaptive system is that it’s extremely complicated to design a very effective adaptive system, both in terms of the platform design but also the content that goes into it.

 

I saw this when my team developed a product powered by Cerego a few years ago. Cerego is a company that has created a “mastery” quizzing platform based on cognitive psychology. Cerego help students “master” concepts by serving questions to them until they are answering correctly all the time. The system also factors in time away from material, showing that “skills” or topics weaken over time, so that if a student goes in only once every couple of weeks, they’ll find their skills have degraded (see screenshot of the dashboard below).

Image caption: The Cerego instructor dashboard can show how students array across a range of levels of mastery. The platform adapts to student skills by repeating content that they struggle with until they show mastery. While students who are more fluent with the material may have to commit less time in a single session, all students must do regular work in Cerego, and students cannot “fail” because they answer items incorrectly--they can only be behind because they haven’t spent the time to reach mastery. (Source: Cerego, Inc.)

 

My team developed quizzing content for a world history course in the Cerego platform. Instructors that we surveyed liked that the platform has gamification elements and is engaging for students. Anecdotally, the Cerego exercises were fun in a way that most history quizzing fails to be. However, there are limits to a system like Cerego. As Kerr points out: “Only if we conceive of education as the transmission of bits of information...could adaptive learning be seen as some sort of solution to an educational problem” (Kerr, 24). In the case of the Cerego implementation, the problem we were trying to solve is students not reading and understanding the basic content of a course (according to instructor feedback) and therefore not being able to move into higher levels of Blooms’ Taxonomy successfully. Students needed fluency with basic content before they could analyze historic events or create historical arguments (in papers or presentations, for example).

 

...However, given a chapter of world history content, what are the foundational pieces of information that students must “master” in order to move on to higher orders of thinking? In language teaching, while the “transmission of bits of information” still may not be the goal, there are at least vocabulary items that students need to accumulate to construct increasingly complex language. In a history course, if the overarching desired learning outcome is for students to think like historians, what events, names, or definitions do they first need to amass? With simple content like that in a Cerego exercise, we run up against the classic accusations about history: it’s just memorizing names and dates or “it’s just one damn thing after another.” This problem is pronounced for history content, in my mind, but it also points out a key issue in learning theory, one that Cope brings up often: is learning memorization or is learning some other ability to produce an artifact calling up many other sources?

 

This also leads us to the next issue, which Kerr summarizes nicely: “Adaptive software can only personalize to the extent that the content of…[a] programme allows it to do so” (Kerr, 10; emphasis is Kerr’s). Regardless of the topic area, a system can only use data to the extent that it is programmed to understand that data. Content must be flexible and robustly metatagged in order for a system to personalize to students. This is an issue but Kerr’s imagination also fails him here. GIven a system like Cerego, if students are getting a definition wrong, all the system can really do is serve that same vocabulary content up in different ways--the system can give them the content as a flashcard, ask if they have it and then serve it as multiple choice in a couple of ways (given the word or phrase, pick the definition/ given the definition, pick the word or phrase)...but the system is not designed to be more flexible. Ideally, perhaps a student could watch a video about the concept or even read or listen to a scenario in which the concept is embedded--the system has only levels not flexible styles of learning.

 

It shouldn’t surprise us that Cerego and similar are not yet fully delivering on the promise of adaptive. It also should be no surprise that even more robust adaptive systems should be designed with some instructor intervention in mind--an instructor should be present and the system should be able to flag when a student is having repeated trouble with a concept or set of concepts. To do this, a system must be fed information about content (meta-information) as well as just the content. Currently, this is somewhere where many platforms are falling down--I have often encountered systems that didn’t really build in robust meta-tagging capabilities, and this means that the ability to make content more flexible simply doesn’t exist in those systems. One issue is that software (not the full learning package built on software) is sometimes seen as the answer--we see this with systems like Cerego that promise good learning outcomes regardless of the content that goes in.

 

At any rate, for the full promise of adaptive to be realized, software/technology and learning must truly work together and inform each other. And this means complex development work that many education technology companies have not yet learned how to do.

 

Sources:

Cope, Bill and Mary Kalantzis. “Chapter 1: Conceptualizing e-Learning.” Last accessed April 1, 2018.

 

Kerr, Philip. “Adaptive Learning.” http://the-round.com/wp-content/uploads/downloads/2014/07/A-Short-Guide-to-Adaptive-Learning-in-English-Language-Teaching2.pdf The Round: July 2014. Last accessed April 1, 2018.

 

Cerego. https://www.cerego.com/

 

Duolingo. https://www.duolingo.com/

 

Openosmosis.org. Video on Adaptive Learning.