Building Personalized Education With AI Adaptive LearningBuilding Personalized Education With AI Adaptive Learning
Building Personalized Education With AI Adaptive Learning
September 25, 2018
By Ciarán Daly
LONDON– Do you remember acing your history syllabus and feeling held back by the rest of the class? How about getting stuck on beginning algebra while your schoolmates raced ahead?
The traditional model of classroom education, sadly, continues to be very much one-size-fits-all. A course syllabus is interpreted and delivered by a teacher en masse to large groups of students, with little flexibility allowed for the bright sparks or the slow learners. This is as true of schools the world over.
However, this one-size-fits-all approach is problematic—especially in highly competitive education systems like that of China. China’s high school and university admissions tests are among the most challenging and demanding in the world, and with every parent looking to get their child ahead, there’s a huge market out there for innovative, individualized e-learning tools.
One such programme, provided by Chinese edtech company Yixue Squirrel AI Learning, goes much further than a simple online course. Boasting over 1,000 education centres across the whole of China, Squirrel AI provides students with AI-powered adaptive learning system which tutors in many different subjects. Working from either a supervised learning centre after school, or via an online space featuring regular video call contact with personal mentors, Yixue tailors a course exactly according to a student’s strengths and progress thus far.
After school, students are able to study in a supervised environment in the education centres or at home. They can work through the course content at their own pace, with each new course item a unique data point of thousands designed to optimize learning efficiency.
Diagnosis & prescription: using AI to adjust content by performance
Any adaptive learning system, explains Yixue’s Chief Data Scientist Dan Bindman, is made possible by three components. The first is a necessary part of all learning: the content itself. Some subjects are better-suited to this than others. Math, for example, has a natural structure and progression to it, whereas more open-ended subjects such as the humanities are more difficult to model. For Bindman, it comes down to the strength of the content itself: “How strong is the content? How complete is the content? How deep is the content? These questions apply whether you’re going to introduce adaptive learning or not: it’s the content you’re going to be teaching.”
Secondly, there’s what Bindman calls ‘diagnosis’. This is where the system identifies exactly what the student does and doesn’t know at an extremely high resolution; something traditionally performed—painstakingly—by a teacher. This is achieved through breaking up the content into units made up of thousands of ‘items’—a group of similar questions that aim to teach students solutions to specific problems.
“If I showed you how to do x+7 = 14 once, I wouldn’t give you that problem again because you’d just do it by memory. Instead, I’ll give you a different problem,” Bindman says. “That makes up a single item, and in all of Beginners’ Algebra there might be 1000 to 2000 items that make up the entire course.”
What makes adaptive learning special is that it figures out exactly which of those items a student is strong and weak at in order to identify what they’re ready to learn. Unlike a traditional classroom where all 30 or 40 students are taught the same thing—regardless of individual progress—adaptive learning is able to provide a highly individualised course syllabus in order to maximise a student’s progress.
“Ideally, you have to adjust constantly to the student, making little incremental changes all the time. Then, at the very end of the course, they’re given a post-cast that measures what their gain was and records their results,” Bindman explains. “Our customers are the parents as much as the kids, and they want to see what the pre-test results were, what the post-test results were, and where the improvements were.”
No future for teachers? Not likely
Squirrel AI’s scope and reach is impressive, but thankfully, Bindman, a former math teacher himself, believes that adaptive learning systems like Squirrel AI and others aren’t about to make teaching professionals obsolete any time soon. Instead, Squirrel AI is designed to support and augment the work of teachers by taking away the need to teach what Bindman calls the ‘nuts and bolts’ of each course.
While YiXue’s content is specifically geared towards the Chinese education system, the successful deployment of the engine itself offers some fantastic lessons for anyone looking to develop their own adaptive learning system.
“When you start with a system like this, there’s two options. One is building AI with limited or no data, and the other is building AI with lots of data,” Bindman says. “It’s much easier to build it with lots of data—it takes time, but you have the data. On the other hand, when there’s not much data, you have to rely on content experts in combination with whatever limited data you have.”
As with any enterprise looking to introduce AI into an organization, there remains the need to mitigate internal cultural obstacles. “If your adaptive system suggests a way for a certain student to learn, and the teacher says they don’t want to teach it that way, then that’s a problem,” says Bindman. “It’s a problem from a business standpoint because it’s taking choices away from the teacher. There’s a need to become more flexible with different teaching styles, because then [the teachers] can appreciate the AI without it disrupting them.”
Moving forward into the future, Bindman sees AI adaptive learning becoming more open-ended and assessment based. The need to deliver content at a deeper level of understanding while allowing for more exploratory learning will become crucial—but this is a huge challenge to assess. The introduction of ‘stealth assessment’, which is able to continuously assess a student’s performance and emotional state using machine learning and other emerging technologies, may one day go some way towards overcoming this obstacle.
Ultimately, Bindman believes that the value of an adaptive learning system comes down to user engagement. “If a student doesn’t work a lot on the product, it doesn’t matter how great theoretically the product is. So how can you deliver that level of engagement without watering it down at the same time?”
Based in London, Ciarán Daly is the Editor-in-Chief of AIBusiness.com, covering the critical issues, debates, and real-world use cases surrounding artificial intelligence - for executives, technologists, and enthusiasts alike. Reach him via email here.