The Knewton Blog

Our monthly newsletter features edtech and product updates, with a healthy dose of fun Knerd news.

“Predetermined Adaptivity”: A Contradiction in Terms

Posted in Adaptive Learning on April 1, 2015 by

Nearly every digital education product today boasts some form of customization. When Knewton was founded in 2008, the idea of personalized learning had been around for a while, but the idea of computer adaptive learning was so new in the marketplace that Knewton’s trademark attorneys couldn’t find a single instance of it in corporate literature (we did eventually find it in academic literature).

Now “adaptive” is an overused buzzword. Products are marketed as “personalized,” “differentiated,” “intelligent,” “adaptive,” and even “super-adaptive.”

Lite Adaptive

Remember those Choose Your Own Adventure children’s books from the ‘80s and ‘90s? They allow the reader to make choices that determine how the plot progresses. But while the book may appear to provide a totally unconstrained world to explore, in fact the reader is forced down predetermined paths into one of just two or three dozen endings.

The structure of many products today marketing themselves as adaptive learning resembles that of a CYOA book. These “Lite Adaptive” products are similarly governed by simple “if, then” rules. For the CYOA book: “If you decide to walk along the beach, turn to page 5. If you decide to climb a rocky hill, turn to page 21.”1 For Lite Adaptive products: “Once the student gets 20 ‘X-Y plane’ questions right, move her on to ‘slope of a line’ questions. Until then, keep giving her ‘X-Y plane’ questions.”

Lite Adaptive products may appear to provide unconstrained adaptive learning. They may claim to be data-driven, or to update in real-time. In fact, they’re just decision trees using if-then rules to herd students in pre-determined directions. And they do so without real data: determining student proficiencies (what students actually know at any given moment in time) is a difficult and expensive process. So Lite Adaptive products instead use general student activity as a rough proxy.

Because they are governed by generalized rules, Lite Adaptive products by definition do not personalize learning for each individual student. The best subject matter specialists in the world can’t identify and create rules for the infinitude of ways a particular student might learn best, or perform differently from others. Instead, Lite Adaptive products lump students together into large buckets with tautological rules like: If Econ 101 students need more practice on supply and demand, they should… do more practice questions on supply and demand! Instructors don’t need a new product for this; they routinely do it themselves in the course of assigning homework.

Building Real Infrastructure

Knewton has taken a fundamentally different approach, creating adaptive learning based on data science, network effects, item response theory, and psychometrics. We’ve built an enormously powerful online platform that provides “adaptive learning as a service” to educational institutions and publishers around the world.

With Knewton, each student’s learning path wasn’t predetermined long ago by a subject matter expert in a cubicle. We think “predetermined adaptivity” is a contradiction in terms. Instead, each student’s path emerges in real time as we learn more and more about the student.

In the seconds before recommending a learning activity, Knewton takes into account — using real data — all of your proficiencies, learning patterns, and past performance. We recalibrate our understanding of exactly what you know, how well you know it, and how you learn it best. Then we examine the content and strategies that worked best for other similar students — using the combined data power of our entire network to find your best possible learning path for every concept you study.

Then, in choosing the exact right piece of content for you, and the exact right way to deliver it, we evaluate every piece of available content to see: 1) how well it teaches a given concept, 2) how engaging it is, 3) how much data it typically generates, and 4) how relevant it is to a student’s current learning goals (as set by the instructor or syllabus). We create the best possible recommendation for every single student for every single concept by maximizing and optimizing these variables. (Take a look at our technical white paper for a more detailed overview.) And we do all of this for every recommendation we ever make — more than 15 billion to date, and growing quickly.

The Payoff for Instructors and Students

As more schools and instructors use adaptive products, they’re experiencing firsthand the stark differences between Lite Adaptive and true adaptive learning. Lite Adaptive products generally just offer recommendations for the next practice question to work on, and do so without real student proficiency data.

In comparison, true adaptive learning solutions, such as Knewton, can today provide all of the below benefits to students and instructors.

  • Practice question recommendations: Recommend the exact assessment item a student should work on next, based on the student’s current proficiencies and academic goals
  • Instructional content recommendations: Recommend the exact piece of instructional content a student should work on next, based on the student’s proficiencies and goals
  • Pinpoint predictions of student performance: Predict how likely a student is to correctly answer a question about a concept, given her current level of understanding
  • “Pre-remediation”: Predict a student’s upcoming trouble spots, and provide targeted content to remediate them before they emerge
  • Differentiate content for engagement: Increase student engagement by differentiating content by pedagogy, difficulty level, and media type for each student, based on what works best for her
  • Optimize content based on student learning patterns: Present learning material in the way that is most productive for a given student
  • Show instructors which students need help, with what material: Provide instructors with powerful data dashboards showing inferred metrics (e.g., a student’s predicted score on an upcoming exam) to help them target interventions
  • Create more effective study groups: Suggest student peer groups that include students with different or complementary strengths
  • Show how well content is working: Show content creators how well every piece of content in a course teaches or assesses students, and point out areas where improved or additional content is needed

These features aren’t a far-off thing. They exist right now. Not all Knewton partners use all of these features in their products, but they are increasingly doing so. Nearly all Knewton partners launching new products in 2015 will take advantage of most or all of these capabilities.

Wayne Gretzky once said: “I skate to where the puck is going to be, not to where it is right now.” The makers of Lite Adaptive products are all skating to where the puck was last year. Adaptive learning is still new, and the market doesn’t yet totally understand it — but it will. Product transparency is quickly improving, too: The next 36 months will bring the greatest increase in the measurability of learning outcomes in the history of the education industry. As instructors and students gain a better understanding of how well different adaptive learning products actually work, Lite Adaptive products will become virtually unsellable. The companies betting their future on simple decision trees need to improve their products dramatically to stay in the game. And that’s great news for students and instructors.


  1. Actual choice from the first ever Choose Your Own Adventure book, Sugarcane Island