What Cognitive Tutors Can (and Can’t) Teach Us About Personalized Learning

Many educators, and especially those interested in educational technology, are currently obsessed with the idea of personalized learning. It’s at the heart of some well hyped initiatives such as the School of One in New York, in which students have tailored schedules, called “Playlists”, that guide them from activity to activity and computer algorithms that generate specialized lesson plans based on a student’s prior performance. The EU’s iClass experiment is also based on this idea of personalization via technology.

The basic promise of personalization is easy to grasp – not every child in a classroom is at the same level, and there’s (presumably) no way for a teacher to teach to all of these differences effectively. In the past, many  educators largely relegated personalized learning to those in need of remediation, the so called “low performers” in a class. This remediation often took the form of tutors, an expensive but effective approach. Later we also saw (and continue to see) Individualized Education Plans (IEP’s, in ed lingo), usually reserved for high risk or special education students. But the idea of having every student engage in some form of personalized learning, evidenced in initiatives like the School of One, may seem unique, but it is not new. Cognitive Tutors, computer programs that aim to replicate the effective guidance and adaptability that human tutors have been proven to provide, have been in the personalized learning game since at least the 1980’s.

Cognitive Tutors essentially incorporate cognitive models of both novice and expert thinking around a certain domain, like math, into a computer program. A learner is challenged to solve a problem that relates to that knowledge with the program providing hints but also taking into account multiple paths towards solving that problem as well as some common misconceptions that are represented in its novice models (Koedinger & Corbett, 2006). Most of the tutors that I’ve seen deal with math and science, and are predicated on the idea that there is one right answer to a problem, though potentially multiple paths towards getting to that answer. Even the most progressive (and impressive) amongst technology of this sort, Dan Schwartz’s Teachable Agents, which flip the model of the cognitive tutor by having the student school the computer as opposed to the other way around, are still predicated on there being one right answer to a particular problem. To me then, I see these as highly sophisticated ways to teach the basics, ie, the stuff that we as a society already know. But what about what we don’t know? Isn’t that the sort of thing that we need to have our future leaders grappling with?

This leads me to what I believe cognitive tutors can shed light on in terms of the model of technology and learning that I’m developing for a course I’m taking, a model I originally introduced and contextualized here and which you can interact with here. I’ve included a static image of the model for reference here:

One of the key innovations that I include in the model is this “Technology Driven Personalization System”, and it’s this idea that I think cognitive tutors can speak to, not because of what they do but because of what they don’t do. The general idea behind this personalization system, for me, is some kind of coordinating body that’s paying attention to all the “nodes” in a youth’s learning ecology and making recommendations for the young person about what might be best to pursue based on that from a learning perspective.

What I’m seeing in the proposal I’m making about personalization here is far less structured than how cognitive tutors conceive of the idea of personalization. It does not assume that there is one “right answer” as to the learning trajectory a learner should follow, indeed, it doesn’t envision an end goal. In contrast to heavily scaffolded learning technologies like cognitive tutors and many games (a technology I’m a fan of from an educational standpoint), what I’m envisioning is much more something that’s about resourcing the young person to pursue their own interests and their own values, as opposed to an imposed standard of what’s important to know. My model assumes that we must trust youth to become active learners, but doesn’t assume that they already have access to the tools and opportunities they need to do so. This is the role of the system I’m presenting here.

At the same time, I acknowledge that every system has its own politics and priorities, and so the question of what kind of  ideology is baked into the system is a very good one. Ideally, what I’d like to see is a system where the inherent ideology is itself  based on the idea of having others bring their own ideologies to the system and ‘make recommendations’ based on them.  Since many teenagers are often not quite at the stage of having very clearly articulated value systems and interests, I can envision the system integrating data about them in multiple ways, some more explicit (profiles with interests they’ve filled out, information about programs they’ve gotten involved with, classes they’re currently taking) and others less explicit (having some sort of match question system, common on dating sites, that don’t directly ask you what you’re interested in or how you think but rather pose situations or hypotheticals for you to respond to that then serve as indicators). All of this would then be integrated to make a profile of a given learner and what they’d like to pursue, which brings us, of course, to the issue of privacy and surveillance.

As someone deeply concerned about issues relating to exploitation and privacy online, my own proposal makes me nervous. Most of us are currently in a situation online where we’re not the customer in places like Facebook, Twitter and Google – we’re the product. Personal data is being packaged and sold to the highest bidder in the form of marketers, and governments are increasingly surveilling their citizens in these spaces. And it’s exactly the kind of personalization and recommendation engines that exist in places like Netflix, Amazon and Facebook, ones based on the existing data about a user, that I would imagine powering a personalized learning system of the sort I’m envisioning. That’s why it makes me nervous, and it’s also why the point I make above about politics and priorities being embedded in the system is so important – given the level of information that something like this would have about a young person it’s essential that it be clearly designed off of the principal of resourcing a young person to pursue their own interests according to their own values.

Finally, I’d envision the system incorporating some of the designs that drive Diaspora*, the open source social network that arose in response to Facebook privacy issues in 2010. In Diaspora, users have full ownership over their data, can share or not share to whomever they want, and simple ways to control privacy are put at the forefront. I would imagine the same, and more, for a system that would have so much data about a young person. And if I truly did believe in the idea of self-determination on the part of the young person, putting them in the position where they were in full control over their footprint within this system would only make sense.

9 Responses to “What Cognitive Tutors Can (and Can’t) Teach Us About Personalized Learning”

  1. 1 amichetti September 27, 2011 at 7:39 am

    Great post, Rafi. Thanks for sharing. I’m intrigued. Are you familiar with the unschooling movement? I thought of it a lot as I read your post. I wonder where your model would fit into that… or would it?

    • 2 Rafi Santo September 27, 2011 at 8:26 am

      Thank Adrienne! I’ve really only heard of the unschooling movement in passing, but doing a quick read on it now it seems *very* aligned with a lot of what I’ve been thinking about recently. Are there any foundational writings you know of on it that I can check out for more *self directed learning*? ;)

  2. 3 amichetti September 27, 2011 at 10:56 am

    Hmm, I’m not an expert, really… I was just curious. I’ve done a lot of reading about it in places like blogs, but no formal academic readings. A good place to start, I imagine, is here: http://www.sudval.org/

    I have very mixed feelings about this philosophy. There’s a big part of me that wants to believe it can work, and there’s another big part of me that things it’s too extreme bc it leaves too much up to the individual and not enough to the community.

    However, I see a “natural” link between their model and what you’ve described in your project… and I find both very interesting. :) Let me know your thoughts.

    • 4 Rafi Santo September 27, 2011 at 12:39 pm

      The school is very interesting, and I can see where mixed feelings would come from. I’m all about interest-driven learning, but we don’t really know what the logical endpoint of an educational philosophy that’s driven by it would look like. I think that if we just switched public school models to the one that Sudbury Valley has overnight, we’d see chaos and a lot of kids wouldn’t know what to do. It really feels like the success of such a model depends on having a strong culture (or cultures?) of active learning that surround an institution that pursue it.

  3. 5 johannakeene September 28, 2011 at 12:37 pm


    I completely agree that the Cognitive Tutors and Teachable Agents are problematic in that there are only certain answers that are deemed right and that anything outside of that is wrong. It seems to me that if we implemented Cognitive Tutors and Teachable Agents we would still have much of the same in classrooms as we do now.

    I am interested to know how you see formal learning that is more interest driven. I really like that idea, but it’s hard for me to visualize what that would look like in a formal classroom. I see this implemented in after school programs and informal learning spaces, but I don’t quite now how I would see that going in a school. What do you think?

    • 6 amichetti September 29, 2011 at 1:06 am


      What you’ve described — formal learning that is interest-driven — is what the best teachers do all the time in inquiry-based classrooms in schools all over the world. It involves a) a teacher who knows his/her students well enough to know what their interests are, and b) knowing how to design learning using backwards-planning (UbD-type stuff, usually). If said students are old enough, they can completely direct their own learning based on interests but still within a formal learning context. My own high school, which I attended from 1989-1993, does this quite well. More about it here: http://en.wikipedia.org/wiki/Bishop_Carroll_High_School_(Calgary,_Alberta)

      Knowing all of this, I think it’s not only reasonable but very viable to imagine how these learning contexts could transfer to online spaces such as what Rafi describes here. In fact, it makes me excited to think of all the possibilities. :)

  4. 7 Asmalina Saleh (@antaera) September 28, 2011 at 8:28 pm

    I’m struck by two things you mention in your post, which are actually two related arguments.

    “In contrast to heavily scaffolded learning technologies like cognitive tutors and many games… [this is] about resourcing the young person to pursue their own interests and their own values, as opposed to an imposed standard of what’s important to know.”


    “Ideally, what I’d like to see is a system where the inherent ideology is itself based on the idea of having others bring their own ideologies to the system and ‘make recommendations’ based on them.”

    In the former, I’m assuming that you are referring to educational games, where the trajectory is scaffolded, than in non-educational games. In fact, a lot of games or online environments seem to align very well with the system that you are suggesting. An environment that comes to mind is Whyville – in some ways, yes, there is scaffolding embedded in the design, but one can also argue that students are able to navigate based on their own interests. In other environments like mmos (LotRo for instance), a lot of theorycrafting draws on knowledge derived formal education spaces and other spaces in your model. From my experiences, all these spaces blend together/overlap and as such, I struggle a bit with the various nodes. For instance, affinity spaces, semi-formal environments and popular culture can intersect and I’m having a hard time parsing out how the distinction can help a student make an informed decision about what to pursue. Indeed, interests can be fluid, and learning does not follow a linear pattern but quite haphazard (at least for me!). How would you address the overlap, or is this one possible outcome? Also, if students have such varied interests, how can we recommend the next steps for them?

  1. 1 Learning Technology for Youth, Learning Technology By Youth « Empathetics: Integral Life Trackback on October 10, 2011 at 9:02 pm
  2. 2 Avoiding the “Creepy Treehouse” When Using Technology to Connect Learning Ecologies « Empathetics: Integral Life Trackback on October 25, 2011 at 12:12 am
Comments are currently closed.

Hi there.

Rafi in thailand, smiling

If you're reading this, then you've reached the web log of Rafi Santo. This is my little slice of the internet where I can share my passion (or whatever) with the world.

Research. Meditation. Learning theory. Spirituality. Activism. Cooking. New Media. Pedagogy. Photography. It's all fair game, and will likely coalesce into some unholy mixture thereof. But hey, that's the integral life.



%d bloggers like this: