Personalized learning – a data revolution

Twenty-first century approaches address the cognitive and motivational state of the learner
Illustration: Mrzyk & Moriceau for BOLD
Illustration: Mrzyk & Moriceau for BOLD

Who doesn’t want a world in which everything they interact with is tailored personally to them?

The advent of big data and machine learning has moved us towards a place where this vision is being realized at an incredibly rapid pace, fueled by our ability to capture and store everything about ourselves – from the keystrokes on our computers to our genetic sequences and DNA. This has led to the rise of personalized medicine that selects procedures that work for our specific body chemistry, personalized advertising that shows us the brand of jeans that fits us best, and even learning personalized to our individual needs.

The immensely powerful idea behind personalized learning is to provide the learner with meaningful learning environments and instruction, as an alternative to a “one-size-fits-all” approach in which everyone in a class sees the same material at the same time. However, the term personalized encompasses a wide range of methods and strategies, with different meanings to different communities, making it confusing to understand precisely how such an approach will be implemented.

“Twenty-first century approaches to personalizing learning are founded on the principle that learning happens by doing rather than listening.”

Some of these approaches are connected to interest-driven learning, in which instruction is personalized by allowing the student to select projects and problems that motivate them, whether it’s the chemistry in environmental activism or the physics behind skydiving. Other methods involve the creation of learning communities that offer small group work and opportunities for teachers to get more deeply involved in understanding the learning needs of their students.

Alternatively, twenty-first century approaches to personalizing learning address the cognitive and motivational state of the learner. These approaches are only possible because of the streams of data that are generated as learners increasingly engage in learning activities online, and are founded on the principle that learning happens by doing rather than listening. Therefore, they focus on practice of important problems in a domain.

AI detects and utilizes strengths and weaknesses of each learner

Personalized learning systems that do this are driven by Artificial Intelligence that models student cognition to assess the current state of the individual learner, and then provides instruction targeted to that state.

This happens in two stages. First, as a learner works on a problem, personalized learning systems detect common misconceptions that the learner has at any step of the way. This allows the system to provide feedback and support that is directly aimed at the issue the learner is having. For example, while working on a fraction addition problem, a young learner might confuse the denominator and numerator. The system can detect this and provide additional feedback and hints that guides them towards an understanding of the numerator and denominator concepts, and also helps them perform the task.

The second approach that these systems take is to select appropriate material for the learner to work on based on what the learner has demonstrated that they know or do not know. This usually involves breaking domains down into the most basic component skills. That way, learners can focus on getting practice in the places they need it most. For example, that same learner may have proven their understanding of fraction addition, but only when the denominators are the same – so the system might specifically target practice of different-denominator fractions for that learner.

“The teacher is freed from the need to homogenize their instruction, and can prowl the classroom to give extra targeted help for those students who need it.”

Personalized learning augments the teacher’s reach

While these systems rely on collecting streams of digital data as students work on a computer, they are typically intended to work within the ecosystem of the classroom. For instance, the teacher may provide instruction in the classroom three days of the week, while students work on computers for the other two. This avoids the problem of the teacher needing to teach to the “average” level of the class.

Rather, students who are ahead and may often feel bored sitting in class feel empowered to move forward at their own pace and frequently become more engaged with the material. Students who struggle to keep up with the teachers’ pace, on the other hand, get personalized assistance from the system. Most importantly, however, the teacher is freed from the need to homogenize their instruction, and can prowl the classroom to give extra targeted help for those students who need it.

“While the idea and initial implementation of such systems have been around for a quarter of a century, now is the right moment in time as advances in deep learning, cognitive modeling, and hardware are converging.”

These systems have been demonstrated in many studies to benefit learners by improving mastery of the content. In fact, some studies have shown that they are particularly beneficial for the students most in need, who would otherwise struggle to keep up with their peers, and more recent work has incorporated additional features that improve learning like virtual agents that motivate students and support for better help-seeking.

While the idea and initial implementation of such systems have been around for a quarter of a century, now is the right moment in time as advances in deep learning, cognitive modeling, and hardware are converging. The era of personalized learning is upon us!

References

Johnson, W. L., & Lester, J. C. (2016). Face-to-Face Interaction with Pedagogical Agents, Twenty Years Later. International Artificial Intelligence in Education Society 2015, 26, 25–36.

Roll, I., Aleven, V., McLaren, B. M., & Koedinger, K. R. (2011). Improving students’ help-seeking skills using metacognitive feedback in an intelligent tutoring system. Learning and Instruction, 21(2), 267-280.

Koedinger, K. R., Anderson, J. R., Hadley, W. H., & Mark, M. A. (1997). Intelligent tutoring goes to school in the big city.