Adaptive Learning | Feature
Page 2 of 4
Adaptive Learning: Are We There Yet?
For more than a decade, K-12 educators have been hearing about the potential of adaptive learning, an approach to instruction and remediation that uses technology and accumulated data to provide customized program adjustments based on an individual student's level of demonstrated mastery. But interest in adaptive learning has been heating up in the last couple of years, thanks to new attention from the Bill and Melinda Gates Foundation, new partnerships among education publishers and adaptive platform providers, and a growing list of product vendors. Along with that increasing interest and expanding vendor landscape has come a fair bit of confusion about exactly what the term "adaptive learning" means. In conversation, it's almost synonymous with "personalized learning," but in practice, these are different concepts, and K-12 districts investigating systems that promise to deliver adaptive learning should understand that difference.
What Is Adaptive Learning, Exactly?
According to Adam Newman, founding partner of Education Growth Advisors (EGA), a strategic advisory and consulting firm and investment bank focused exclusively on the education sector, "'Personalized learning' is really an umbrella term." In two recently published white papers commissioned by the Gates Foundation ("Learning to Adapt: Understanding the Adaptive Learning Supplier Landscape" and "Learning to Adapt: A Case for Accelerating Adaptive Learning in Higher Education"), EGA researchers — Newman among them — defined "personalized learning" as a "pedagogical method or process that draws on observation to inform tailored student educational interventions designed to increase the likelihood of learner success." As Newman said, technology isn't actually required for personalization, but the tech makes it possible to personalize at scale.
K-12 educators have been personalizing learning in their classrooms for decades without technology: If Jesse is having trouble reading, the teacher assigns her some extra reading in Chapter Two, for example. Personalized learning covers a range of approaches and models, Newman said, including competency-based learning, differentiated instruction and tutorial models — as well as adaptive learning. "So you can think of adaptive learning models as one approach along a spectrum that enables personalization," he said.
In "Learning to Adapt," EGA researchers went on to define "adaptive learning" as an approach to creating a personalized learning experience for students that employs "a sophisticated, data-driven, and in some cases, nonlinear approach to instruction and remediation, adjusting to a learner's interactions and demonstrated performance level, and subsequently anticipating what types of content and resources learners need at a specific point in time to make progress."
Types of Adaptive Learning
The EGA researchers concluded that technology vendors offering truly "rigorous adaptive learning solutions" leverage numerous areas of academic research, including intelligent tutoring systems, machine learning, knowledge space theory, memory and cognitive load theory. They also divide adaptive approaches into two categories: "facilitator-driven," which refers to products that provide instructors with actionable student and cohort profiles — essentially dashboards. This approach is content-driven, which means that the dashboard output links a specific course's content inventory "within a system of standards or learning sequences."
The other approach — the one most people are talking about when the conversation turns to adaptive learning — the researchers call "assessment-driven." In this approach, the system provides close-to-real-time (sometimes called "dynamic") adjustments of the instructional content. Facilitator-driven systems provide instructors with information they can act on; assessment-driven systems make their own adjustments. In order to provide these adjustments, the researchers said, assessment-driven systems must be correlated dynamically with assets, items and learning objects to standards, outcomes or other frameworks. Another essential difference: the assessment-driven model allows students to move the course individually or in a group, without instructor interaction. Newman added that the two approaches are not mutually exclusive, and both might be found in a single product or system offering.
Partnering To Adapt
One recent trend that promises to advance adaptive learning in K-12 significantly is partnerships among educational publishers and adaptive learning platform providers, such as CogBooks, CCKF and Knewton. In 2014, for example, Knewton announced a partnership with Microsoft, which will make the Knewton API available to "its vast partner and publisher ecosystem." And in 2013 K-12 educational content company announced plans to build new digital products using the Knewton API. Knewton has also been working with Triumph Learning and Houghton Mifflin Harcourt (HMH) to add adaptivity to their K-12 learning products.
Educational content provider Triumph is using the Knewton API to enable adaptivity in new digital products aligned with the Common Core. Triumph is set to launch these products this year in multiple subject areas. HMH worked with Knewton in a previous collaboration on a project to provide adaptive learning in math, science and language arts to incarcerated K-12 students. HMH is now using the Knewton platform to bring adaptive learning to its mainstream K-12 content in math, reading and other core subjects. The publisher is currently in the process of adding the Knewton-powered adaptivity to its GO Math! program.
Adaptivity in Action
Mary Cullinane, chief content officer and EVP of corporate affairs at HMH, said, "There has been a lot of noise in the marketplace around adaptive learning, but so many of the offerings are just binary 'if you get this question wrong you go here; if you get it right, you go there' kinds of things. True adaptivity isn't just about understanding that the kid got the question wrong, but why the kid got the question wrong."
The Knewton platform, for example, is a set of adaptive learning infrastructures designed to allow third-party vendors to add adaptivity to their products. The platform collects and processes data from real-time streams and maps the relationships among individual concepts within the learning content, which are then integrated into taxonomies, learning objectives and student interactions. It then uses this data to evaluate student proficiencies and generates "insights" and predictions that lead to recommendations. The goal is the creation of an individual "learning path."
David Kuntz, vice president of research and adaptive learning at Knewton, explained, "Our partners are the experts in their target market. They create the application and pass us the data. We process that data and make a set of actionable inferences about the students, and then pass those back to the application, and the partner decides how and when to render those for the student."
In this clip produced by Knewton, the company's content developers explain how they map learning concepts to prerequisite skills in order to create personalized learning paths. (Video courtesy of Knewton)
HMH is currently taking its adaptivity-enhanced math program for a test drive with more than 2,000 students in California, but the company is keeping the details of that "closed pilot" test under wraps for now. "As an industry we tend to overpromise," Cullinane said. "Everybody starts using 'adaptive' or 'personalization,' but the truth is, we're still in the early stages, and we want to be super-honest about that. But we do feel that this is technology that can actually deliver on its promise. It's going to take a little time to make sure that we do it right."
The Holy Grail of Data
Data collection is essential to any adaptive learning system, of course, but according to Kuntz, such a system applied in a K-12 environment presents a unique opportunity. "One of the great things about K-12 is that students spend a long time there," he said. "You get to collect data from students year after year as they progress through the curriculum. That kind of longitudinal data could enable a view of the student that no one has today. You get to see them in all of their classes with all of their teachers and all of their proficiencies, what they're prepared for and not, what kinds of things engage them and how that develops over time."
"You could call it the holy grail," Newman added. "It's this idea of being able to build a learner profile, such that the instructional pathway the student pursues over his or her lifetime is being dynamically modified and adjusted based on a whole bunch of data and information."
"The truth is, we're just scratching the surface," Kuntz said. "When we start to have the really long-term longitudinal data — where we start to see students from early in their educational careers into adulthood — that's when we will be able to gain truly deep insights into how people actually learn and how they develop over time. We'll be able to discover, say, whether that thing you learned in third grade actually mattered, in terms of your preparedness for higher-order concepts later on. We'll start to be able to see learners in a way we've never been able to see them before. That picture of each individual student and how it changes over time will be invaluable."
That data is especially valuable to the content providers, Cullinane said. The more a student interacts with an adaptive product, the better that provider can steer students in the right direction over time. "Remember, I'm not just looking at one student," she said. "I might be looking at 10,000 students for whom a particular piece of content is working. When I see the 10,001st student with the same set of characteristics, I have a much better chance of pointing that student to the appropriate content. I guess you could say that, as the platforms become more adaptive, the content is getting smarter."
A Premise, Not a Panacea
In K-12 education, the adaptive systems undergirding that content also have to be sensitive to noncognitive data, because students are undergoing cognitive development as they progress through the grades. This makes an "assessment-driven" approach to adaptive learning especially helpful. As the EGA researchers pointed out, the more sophisticated solutions "model and categorize learners through the aggregation of cognitive and non-cognitive data, resulting in a more three-dimensional 'profile' of the learner." The ability to generate this learner profile is what enables a solution to personalize a learner's experience. As the researchers wrote, "This approach requires the greatest degree of technical acumen, as the system must monitor, track and analyze extensive, large-scale data ranging from previous learners' experiences to cognition, modalities and social learning, among others."
This observation points out an aspect of the emerging adaptive learning landscape that might not be immediately apparent: There are varying levels of adaptivity; a product or platform isn't simply adaptive or not adaptive. The differences, Newman said, boil down to understanding what the end goals or learning objectives are, and the level of granularity with which those goals are captured.
"Let's say we're talking about a student's path through Algebra I," he said. "Some of the courseware products will give that student a diagnostic that says he needs to work on quadratic equations. But those products might not know what it is about quadratic equations that's challenging to that student, personally. More sophisticated approaches we're seeing now seek to provide an understanding that it's not just that the student doesn't understand quadratic equations, but that it's these three steps in solving for quadratic equations that really trip him." Then the focus becomes building the student's mastery of those steps, not just quadratic equations in general.
To get to that level of granularity, an adaptive system must break down the learning into building blocks. The more sophisticated systems understand which blocks are prerequisites to larger concepts. The current crop of adaptive solutions that get to this level of granularity, Newman says, are mapping to a set of known, required objectives or competencies.
"Whatever the approach," Newman said, "the essential premise of adaptive learning — and it is a premise, not a panacea — is that it holds the promise of a more outcomes-oriented system that is more efficient in the use of time and resources." He concluded, "Adaptive learning is one strand of personalized learning, and one way to imagine the model, that holds a lot of promise. It can be powerful, but there are lots of options that an organization can explore."