There’s a scene in “Minority Report” you might remember that’s meant to depict the future of advertising: as people walk through a crowded mall, banners scan their retinas to beam them targeted sales pitches. While retina-scanning is extreme, this kind of customization doesn’t seem like science fiction anymore.
Almost every aspect of our day-to-day lives can be personalized. A world where radio and magazine ads adjust to each individual listener or reader isn’t so hard to visualize if you’ve ever looked at the sidebar of a Google search — or signed up for an account on Amazon or Netflix.
Services like Netflix are renowned for their “recommendations” – that is, their ability to evaluate your prior purchases and make conjectures as to what you might like to buy or watch in the future.
As with Knewton’s adaptive learning engine, there is a fair amount of data mining involved in making these recommendations: Netflix’s systems weed through millions of potential recommendations to find you movies you will likely enjoy based on your history. In a similar fashion, Knewton’s Adaptive Learning Platform uses your responses to test questions to make intelligent recommendations regarding content and format to help you master material.
In fact, you could think of Knewton as a sort of Netflix for learning, if that helps. However, the analogy has its limitations, since the task of education has to account for something more complex than preferences. Netflix and Amazon predict that you will like or want a film or book based on what you already like, or what you have already bought. At Knewton, we have to predict that you will learn from something based on what we know about how you learn.
Getting personal with statistics
All recommendation engines attempt to solve problems by making statistical inferences. Amazon uses your purchase history, making recommendations based on what people with similar histories to yours have gone on to buy. Netflix uses your ratings and rental patterns to predict what you might rate highly next.
Hunch is a service that takes a slightly different approach, creating a broad personality profile for its users to make recommendations across categories. Instead of suggesting a book based on your reading a similar book, it might recommend books, magazines, or TV shows based on your responses to a number of questions.
This is closer to what the Knewton platform has to do; instead of personality profiles, though, we’re interested in building learning profiles. And the Hunch approach doesn’t quite capture something that an education engine has to. To be of any use to students, adaptive learning also must account for both the hierarchy and format of the material it recommends.
Good teachers know that lessons need to be taught in a logical order, with each new concept building upon a prior understanding. What’s more, how a concept is presented is critical, as some students do better with lectures while others need visuals or interactive exercises.
Hunch, Netflix, and Amazon make excellent recommendations based on what they know about their users. For adaptive learning to work, though, we have to tackle a slightly different challenge.
Imagine if Hunch could not only tell that you would be interested in the story True Grit, but also could predict that it would have a bigger impact on you in book over movie form. Imagine if Netflix could not only recommend the films of Akira Kurosawa, but also determine in which order you should watch them to best appreciate his work.
Netflix doesn’t take the order of a viewing experience into account because it doesn’t need to, but things like lesson and concept progression are essential to the task of adaptive learning. To make personalized learning suggestions that actually work, an education engine must do more than simply gauge your taste; it needs to understand the way your mind works and have an appreciation for your ultimate goals.
This is why we’re so excited about adaptive learning. By taking into account student profiles, concept relationships, and learning modalities, an adaptive platform can provide uniquely personalized feedback to students — which in turn can help them reach better learning outcomes in a fraction of the time.
On a student level, this means that every learner gets precisely the material he or she needs, customized to his or her learning style. Students who aren’t being challenged get more difficult questions; formally “unteachable” students get targeted help, allowing them to finally master the skills they need to succeed. This may not be as sexy as Tom Cruise’s computer—but ultimately, we think, its impact will be much more significant.