The RAND report also mentioned a point that most of us already knew—there’s not an agreed-upon definition of personalized learning. I see the term used to highlight three possible features of a learning environment. These three are not mutually exclusive.
- First, it might represent a tailoring of pedagogical methods to individual children—one child might learn best like this, whereas another child learns best like that. “Personalized” means “personalized pedagogy.”
- Second, it might mean allowing children to learn at their own speed. There’s no differentiation of pedagogical strategy, but kids who understand move on, and kids who need more instruction get it. “Personalized” means “personalized pace.”
- Third, it might mean different content for different students, depending on their interests. “Personalized” means “personalized curriculum.”
Any or all of these seem worthy to me, but each carries a challenge that I’ve not seen discussed.
Personalized pedagogy. I can see two ways of doing this: theoretically driven, and theoretically agnostic. In the theoretically driven scheme, we have a theory in mind about different types of learners. For example, when learning an abstract idea, maybe some people learn best by exposure to many concrete examples before you hit them with the abstraction, whereas others learn best if they see the abstraction before the concrete examples. So education improves if we (1) have in hand distinctions like this that make a difference to learning and (2) have a reliable way to putting kids in the right category.
The problem is that we don’t have reliable theoretical distinctions in hand. (And yes, learning styles would be a subset of this idea.)
The theoretically agnostic version might work this way: you don’t pretend to know the differences among students. Instead, you let a learning algorithm figure it out for you. You note which lessons a child learns more quickly or more slowly, and keep a running tally. Over time, you should see which type of lesson each child learns more quickly. Then you can give the child that type of lesson more often (presumably still varying them some, so that you can continue to fine-tune your understanding of the child’s preference.)
This method is actually not theoretically agnostic. You still need to pick features that you’ll code for each lesson. The number of features we might attribute to each lesson is pretty big. That is not a problem for the learning algorithm, but may be a problem for creating lesson plans. The larger the number of features I code, the more likely I am to capture a set of features that’s a good fit to an individual student. But the larger the set of features, the longer it takes me to sample the feature space (i.e., the more lessons I need to administer to get an idea of what’s a good fit for the child).
But the biggest problem is that I’m greatly increasing the number of lessons I need to have at the ready. If you learn math best in a series of 5 brief lessons, with lots of ducks used as examples, with frequent review of previous concepts, and with the use of spatial metaphors, whereas I learn math best in a series of 3 slightly longer lessons, with examples from the solar system, and a moderate amount of review of previous concepts, and the use of number line metaphors. Now suppose everyone in the class has their own set of preferences. How are these specialized lessons going to be generated?
Personalized Pace: The challenge here is similar to the last point made about personalized pedagogy. In this plan we’re not thinking that different children will receive pedagogically different lesson plans. But, if you understand a lesson and I don’t, I will get another explanation, another set of problems to work, something. This decision to offer more instruction and support to me must be based on some decision about my performance to that point. So we have a bunch of decision points where kids either move to new content or review old content in a different way. As we add more of these decision points we have greater and greater opportunity to adjust the lesson based on the students current understanding.
What people often fail to realize is that each decision point also demands new material. A new explanation. A new metaphor. A new set of problems to work. With more decision points, the pathway through the lesson gets “bushier” and “bushier” and we greatly multiply the amount of high-quality instructional content we need. That’s a formidable challenge.
Personalized content: What if we allow students a greater voice in selecting the content that makes up their education? I’m ready to believe that there could be a benefit to student motivation, and that for some, that benefit could be significant.
But I also see a trade-off. Students will go deeper on X, and will slight Y. When student choice is to affect curriculum, advocates tell skeptics that breadth will still be assured. Choice represents a bonus, an “and.” I’m doubtful. I think it will be an “instead.” Even curricula with the explicit goal of breadth, of minimum competence in all domains, struggle to achieve it. If significant time is siphoned off for a particular domain, proficiency in others will suffer.
Practice matters. Students need time to work with ideas to really absorb them, make them part of their thinking. Even minimum competence requires exposure and thinking over the course of a few years.
If students become more narrow—that is, really good at what they are interested in, and less good at what they are not interested in--that’s not intrinsically bad. It’s a choice. It’s a way to reify an educational value. Your goal can be breadth or your goal can be depth. The personalized content approach is a depth approach. We should not kid ourselves that personalized curricula will give us both.
So, all in all, is personalized learning worth pursuing?
I’d rather start with changes that research gives us more confidence will help kids. But hey, that’s just my personalized point of view.