The future of adaptive learning as an iPhone

Dan Meyer in Adaptive Learning Is An Infinite iPod That Only Plays Neil Diamond draws a line between futurists and educators. Futurists envision adaptive learning technologies that replace teachers who fail to give complete individual student attention and enforce a uniform classroom experience that abandons students who are behind and bores students who are ahead. To Meyer, this technology will necessarily lose a lot too: the richness that happens in a live, simultaneous classroom experience.

I don’t yet concede that all will be lost. My aim with this post is to understand the learning benefits of a good classroom that Meyer sees in order to provide suggestions to future software designers (whether or not they adopt the “futurist” label) to preserve and even enhance these benefits. As we will see, there’s hope for adaptive learning beyond Neil Diamond and even the infinite iPod. My model of classroom learning may be incomplete, but then I hope you’ll be able to point to what is missing and somebody (that is, me) will have learned something.

The first thing we think of in rich learning is content. Content, at a pure informational level, can largely be carried over to a digital adaptive learning system:1 record a video lecture of the teacher saying the same words, for example. The popularity of the flipped classroom attests to that.

The first design imperative is to seek out effective educational content for learning systems, then to understand how its audience responds and react accordingly (as a good teacher would).

When we say that richness comes not from the informational content but rather from the presence of a live teacher or peers, that isn’t so much richness of the content as it is of environment. This is what Meyer is referring to when he talks about classroom-based math education

…as a social process where students conjecture and argue with each other about their conjectures, where one student’s messy handwritten work offers another student a revelation about her own work, a process which by definition can’t be individualized or self-paced…

Meyer wants to preserve the liquid networks (Where Good Ideas Come From) that are peers engaged in common learning tasks. Better ways to get from a student’s current mental state A to a better-learned state B may come as flotsam from a peers who is approximately around A rather than from the teacher who is well-accustomed to B. Or from computers that lack any empathy that isn’t preprogrammed. In Dear Teachers, Khan Academy Is Not for You I talk about how the fact that Sal Khan’s perspective may, in some cases, be closer to the students’ mental states than the teachers who criticize the video.

By preprogrammed empathy, I mean that computers can respond to “errors” that it knows about, and may have an excellent approach to help the student correct that error. As computer-based learning scales, it can start to learn more than a teacher about the best directions from A to B, and it can give those directions with complete patience and without falling back to the B perspective too quickly.

This leads to what I think is the ultimate battleground for classroom versus computer learning, feedback. On one hand, a computer’s feedback can be instanteous and adapt the entire learning experience accordingly. Meanwhile the teacher will grade your paper in a week, and though she’ll realize you didn’t understand any of that stuff, there won’t be time to change the lesson plan. But can computers match the targetd and contextual feedback that humans can give?2

Feedback can take on many forms:

  1. Correctness feedback. Software that can evaluate the correctness of something can easily provide right/wrong feedback. It seems that mere correctness feedback can do a lot for learning, but that is an argument for another post. However, computers have a huge artificial intelligence barrier to cross in terms of being able to evaluate what people learn except in limited formats.

  2. Content resequencing. The next step beyond stating whether a student’s work is correct is to adapt the content in response. This can be as simple as repeating the exercise set if a threshold is not reached, as DuoLingo and Khan Academy do. But it can extend to recognizing the details of what is being missed and presenting more instructional content.

  3. Environmental affordances. Beyond the people in it, a classroom environment isn’t particularly well designed for learning. As I talk about in a comparison of learning environments with the game Portal, we can do more in a virtual environment to directly benefit learning. The environment itself can shape your understanding of errors in your thinking and paths to correct them. For example, a tall ledge dropping off in front of you affords figuring out another way to use your portal gun. This idea goes well beyond physical affordances, as I’ll talk about in an upcoming post.

  4. Dialogue. I love the quote from John Holt’s How Children Learn: “To rescue a man lost in the woods, you must get to where he is.” Another Meyer post convinces me of the power of a teacher’s response within the rich context that is the student’s own thinking. For example, a girl is solving a problem that states that 1 in 3 families own dogs and asks how many students in her class may own dogs. The student draws lines for each student in her class and underlines every third. A teacher can recognize that the student is primed to represent the problem as division and can work with the student’s current representation to do that (maybe, I’m not a teacher). That is hard for a computer.

Overall the state of computer feedback is inconclusive and presents a vast opportunity to make computers smarter both in recognizing student mental states and helping them transition to better ones. We have seen research results of adaptive learning systems providing significantly better learning, but the nature of control groups in these studies don’t necessarily imply computers are anywhere close to the best of classroom learning.

If we remain optimistic though, adaptive learning can be not just an iPod that plays any kind of music, but an iPhone where we can program it to do almost anything. This is obviously true: the iPhone and adaptive learning systems are both just computers. Better yet, I hope that a adaptive learning platform can mirror the platform of the iPhone (which includes physical convenience, UI standards, inputs like voice and camera, etc.) that support a beautiful diversity of apps. Apps here being learning experiences that are rich in environment, content, and feedback.

The better analogy is do you want an MP3 or do you want live music? As far as music goes, the world has chosen. Both!

1 Technology skeptics have some basis for distrusting what translates to a screen. Humans have to learn to learn from screens, rather than other humans. For example, infants don’t pick up a second language as readily from a multimedia program as they would from a nanny. But we do learn to use–and seem to fully embrace–digital learning. Many studies have confirmed the engagement of children with virtual entities. Try one yourself: watch a kid play a videogame.

2 There is a middle ground between pure human feedback and pure computer feedback. Computers can provide hints to the teacher about the context in which to provide individual feedback. However, current solutions are not very good, so this is yet another design challenge for educational technologists.