Adaptive Learning Is An Infinite iPod That Only Plays Neil Diamond

I was in a small room recently with some futurists who were very excited about adaptive learning. The reasons for their excitement wouldn’t surprise you. “Prussian factory model of learning, learn at your own pace, et cetera.” I admit it all sounded very appealing and when I tried to articulate my frustration with their model, I didn’t get far at all. I sounded like just another rent-seeking teacher trying to preserve the outdated model that cuts his paycheck.

Futurists and math educators talk past each other. If I could jump into any futurist’s head and encode any particular understanding there to make dialog easier, it would be this:

Adaptive learning is like an iPod with infinite capacity and infinite capability to play any song ever recorded or sung, provided those songs were written by Neil Diamond.

If all you’ve ever heard in your life is Neil Diamond’s music, you might think we’ve invented something quite amazing there. Your iPod contains the entire universe of music. If you’ve heard any other music at all, you might still be impressed by this infinite iPod. Neil wrote a lot of music after all, some of it good. But you’ll know we’re missing out on quite a lot also.

So it is with the futurists, many of whom have never been in a class where math was anything but watching someone lecture about a procedure and then replicating that procedure twenty times on a piece of paper. That entire universe fits neatly within a computer-adaptive model of learning.

But for math educators who have experienced math 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, computer-adaptive mathematics starts to seem rather limited.

Lectures and procedural fluency are an important aspect of a student’s mathematics education but they are to the universe of math experiences as Neil Diamond is to all the other amazing artists who aren’t Neil Diamond.

If I could somehow convince the futurists to see math the same way, I imagine our conversations would become a lot more productive.

BTW. While I’m here, Justin Reich wrote an extremely thoughtful series of posts on adaptive learning last month that I can’t recommend enough:

Featured Comments:

Kent Haines:

Can I offer another analogy for these technologists? Adaptive learning is like a guitar teacher who teaches you how to play harder and harder pieces of music but never teaches you how to improvise. So you can play a piece of music that is placed in front of you, but you’ll never be able to pick up a guitar and just play with a couple of friends. I would contend that the improvisor is better prepared to understand and even make music. I’ll bet Neil Diamond can pick up a guitar and jam.

About 
I'm Dan and this is my blog. I'm a former high school math teacher and current head of teaching at Desmos. He / him. More here.

29 Comments

  1. Futurist are always funny. Great ideas for which they have no working concept. Sound cool, don’t work in the real classroom. It is fun to listen to futurists and watch the crowd. All the prospective teachers that have never taught are taking notes like mad. All the experienced teachers are sitting there with funny grins thinking “Put this guy in a classroom and watch the fun.””

  2. Or, math can be like Netflix and you learn about all sorts of things that you normally wouldn’t get from a typical school curriculum that’s set to cater to the mass audience of kids. Enjoy those mindless blockbusters and see what great kids you turn out. The instructor is just as Neil Diamond as the iPod analogy except with the iPod you can pause, skip rewind or jump to another song if you already know the one playing.

  3. Timely post. I wouldn’t normally plug this here, but I’m actually kicking off an Adaptive Math Learning community next week with a webinar about separating edtech fact from fiction with adaptive learning (http://www.edweb.net/adaptivelearning). I think this is a critically important conversation because it helps focus the discussion on pedagogy and outcomes.

    It’s important to distinguish between (1) buzzwords used by non-educator futurists and (2) learning principles validated by educational research. I was at a talk given by Hugh Burkhardt last year where he discussed how teachers need to have an “adaptive expertise.” So the idea that classrooms and software should both be adaptive is a reasonable expectation and a promising opportunity to improve the understanding of all students. It’s not just a “nice to have.” It’s hard to imagine how classrooms, apps or software could truly be effective for learning if they don’t have elements of adaptivity in which students get useful feedback on their thinking, strategies, and solutions. That feedback might come from teachers, peers, or technology.

    Certainly what teachers or software are adapting to is what’s critical. The key in classrooms and in software is engaging students with meaningful tasks and problems to think critically about and work toward making sense of and solving. Adaptivity by people or computers doesn’t mean much if the tasks are narrow and shallow or if the problems don’t invite students to use their own ideas in ways that surface misconceptions and deepen conceptual understanding. But as you rightly point out, too often adaptive learning programs just use lectures and drill practice.

    Software can’t (and shouldn’t try) to do everything; students should be collaborating on rich tasks with each other. And well-designed software can support student learning and complement these classroom experiences in ways that weren’t possible without digital technology. In 2012, I wrote about how pedagogy must drive the design of adaptive learning platforms (http://bit.ly/10IEAKu), and I went into more depth about IPI and other adaptive learning concerns. That post also explains why I’d suggest this post be titled “Some” adaptive learning is like an infinite Neil Diamond iPod.

  4. It’s hard to imagine how classrooms, apps or software could truly be effective for learning if they don’t have elements of adaptivity in which students get useful feedback on their thinking, strategies, and solutions.

    What feedback is useful for student learning? How does it lead to student learning? Could we train a computer to give feedback like this?

  5. Michael – I’m convinced that we can.

    Computers can already identify facial expressions. At some point, they’ll be able to measure your confusion and stress through blood pressure readings and cameras.

    Computers can already interpret the spoken word (hi, “Siri”). At some point, you’ll be able to talk your computer through a given problem – and it will notice when you’re excited or unsure.

    When we’ve reached that stage, a carefully chosen algorithm could press the student, offer help – or simply signal for the teacher. All these issues could be placed in a web, and eventually the computers will be spot patterns the teachers could never see – like did you know that Johnny in class 2S and Kate in class 2B have trouble with exactly the same problems?

    It won’t be tomorrow, it won’t be easy, and it shouldn’t replace the human teacher (though it might mean different class structures – imagine a class of 100 students with 4 teachers leading tailored workshops, while the other students work with their computers). But the price really is enormous.

    To me, the potential for transformed learning is all the more reason that Dan should challenge the limited vision of so many futurists today.

  6. I think that this challenge you articulate is the most important challenge for educators of our stripe: How do we explain constructivism to an audience that has never taught?

    Can I offer another analogy for these technologists? Adaptive learning is like a guitar teacher who teaches you how to play harder and harder pieces of music but never teaches you how to improvise. So you can play a piece of music that is placed in front of you, but you’ll never be able to pick up a guitar and just play with a couple of friends. I would contend that the improvisor is better prepared to understand and even make music. I’ll bet Neil Diamond can pick up a guitar and jam.

  7. Agree that technology doesn’t replace these social and interactive processes. The trouble is, (a) I don’t think anyone is seriously saying it does, (b) there are contexts where students _want_ to pursue self-study, and thus where something adaptive is better than something that isn’t.

    If you think of adaptive learning tools more in terms of materials than lectures, the conversation becomes more about what happens outside the classroom — during homework, during study hall, etc. Students spend a lot of time with a book, or with the digital equivalent. Those materials can do more than just sit there.

    In addition, talking about “adaptive learning” in the abstract while pinning it to a subject and grade level doesn’t do the discussion any justice. It’s necessary to be specific about where dynamic materials and better insight for teachers/parents/tutors can and can’t help. Every use case involves a specific subject, grade level, product manifestation, and the implementation that surrounds it (faculty, company support, etc.). You’ll see stark differences in how companies implement it with, for example, adult ELT environments vs. higher ed biology vs. middle school math vs K-3 anything.

  8. @Kent — The “improvise” analogy is better. While the Neil Diamond iPod points out an inadequacy, it does not distinguish the “adaptive learning” paradigm from one that also encompasses constructivist techniques.

  9. Any kind of instruction can be done well or badly, and adaptive learning is no exception. But adaptive learning, when done well, utilizes constructivist techniques, such as active learning, guided discovery, modeling, scaffolding, etc. (Sure, they have different terms, but functionally the methods are the same.) Well done adaptive learning also allows for not just training of defined skills, but also for the extension and generalization of those skills to novel situations….”improvisation,” if you will. But in order to be capable of improvisation, a learner needs to be accurate and fast (i.e., fluent) with the component skills.

    I completely agree that there are numerous examples of adaptive learning done badly. Just because someone understands algorithms doesn’t mean they understand what learning questions need to be asked, and vice versa. The challenge for edtech companies who are doing adaptive learning is finding team members who understand both. For a very well done example, I suggest checking out the Headsprout online reading programs: http://www.headsprout.com.

  10. If you think of inquiry lessons as this 2-stage process:
    1). Inquiry
    2). Discussion –> lecture –> practice

    …then adaptive learning seems useless for the deep thinking part. But I think inquiry works best like this:

    1). Inquiry
    2). Whole-class debrief
    3). Back to inquiry: can you do it again under slightly different circumstances, or translate the discovery you made to a different representation, or in some way consolidate your discovery into something you’ll remember.
    4). Discussion –> lecture –> practice

    Then adaptive learning could be useful in Step 3. I wish Desmos would do some of that–they’d do it very well.

  11. Tim Hudson:

    I was at a talk given by Hugh Burkhardt last year where he discussed how teachers need to have an “adaptive expertise.”

    This isn’t Burkhardt’s term, as I’m sure you’re aware. He’s referring to Hatano’s concept of “adaptive expertise,” as distinguished from “routine expertise.” I don’t think this is a bar adaptive learning software is close to clearing.

    From a paper by my dissertation advisor:

    Hatano and Inagaki (1986) described several qualities of adaptive expertise that distinguish it from routine expertise. These include the ability to verbalize the principles underlying one’s skills, the ability to judge conventional and non-conventional versions of skills as appropriate, and the ability to modify or invent skills according to local constraints.
    Wineburg (1998) and others (e.g., Bransford & Schwartz, 1999) have added to this list by pointing out that adaptive experts are also more prepared to learn from new situations and avoid the over-application of previously efficient schema (Hatano & Oura, 2003).

    What is a adaptive learning right now if not a collection of efficient schema?

    Robbie:

    Agree that technology doesn’t replace these social and interactive processes. The trouble is, (a) I don’t think anyone is seriously saying it does, (b) there are contexts where students _want_ to pursue self-study, and thus where something adaptive is better than something that isn’t.

    I can’t think of many futurists who have bothered to articulate the limits of the adaptive technologies they propose. I’m articulating one limit here: proof, which is very difficult to do in a self-paced, adaptive environment. If there are loads of futurists making similar disclaimers, I’ve missed them.

    Karen:

    Any kind of instruction can be done well or badly, and adaptive learning is no exception. But adaptive learning, when done well, utilizes constructivist techniques, such as active learning, guided discovery, modeling, scaffolding, etc. (Sure, they have different terms, but functionally the methods are the same.)

    My point here isn’t that instruction can be great or lousy whether the teacher is a computer or a human being. My point is that some kinds of instruction just aren’t possible (in 2014) when the computer is a teacher.

  12. @Dan, re: your comment to Karen directly above, isn’t your point that some kinds of instruction won’t be possible in the (near-term) future, as long as the computer is the teacher? We are talking about futurists, after all.

  13. Scott Farrand

    May 25, 2014 - 6:49 pm -

    If I get to talk with futurists who are excited about adaptive learning, I might ask them to bring it, and create an adaptive learning system that produces skilled futurists. If this challenge seems unfair, then let’s examine why. How is it that becoming a futurist (and I am not at all certain what that is) is really different from becoming a mathematician? Maybe those aspects of my proposed futurist development system that would seem difficult to program are analogous to areas where adaptive learning currently fails so miserably in mathematics instruction.

    I confess to being cranky on this subject, because the adaptive learning systems that I have seen have done quite a disservice to mathematics learning. These systems are surprisingly successful at producing students who can get correct answers on standardized tests, but in my experience they are a disaster in producing understanding of math. I often teach students who have been taught via adaptive learning software, and the lack of correlation between their test scores and their mathematical sophistication can be very surprising. These systems serve to promote students into classrooms where they are at a significant disadvantage, because their knowledge is situational and their understanding thin.

    Perhaps there are better days ahead for adaptive learning of mathematics. Thus far the strategy of collecting a detailed list of math skills and then structuring the system around progress through those skills has managed to fundamentally miss the fabric of mathematics. I’d suggest that the futurists pay careful attention to the unintended effects of the current systems before they extrapolate.

    BTW, I love the analogy. I suspect that there was some amusing thought leading to the choice of Neil Diamond over other musicians.

  14. Robert Catanuto

    May 26, 2014 - 12:17 am -

    Truth always sits in the middle.
    Neither entirely ditching adaptive learning software is good nor adopting it all the time.

    I’ve seen myself with students many cases where adaptive tools provided a boost in learning, and other situations where engaging in productive discussions among peers and with teacher were a blast.

    For example, why don’t use adaptive learning software as a means for discussion between students and teacher?

    Hope this helps.
    Rob

  15. @Dan — In that particular talk, I don’t recall Burkhardt referencing Hatano. Three slides after mentioning ‘adaptive expertise,’ he discussed design tactics for several teaching challenges, the very first one being that “1 teacher has to adapt teaching to 25 students.” It’s this classroom reality that I was referring to in my first comment and that I think is most relevant to this iPod conversation. It’s good to have a thorough definition of ‘adaptive expertise,’ and I think we need to better define what ‘adaptive learning’ means. I don’t think there’s agreement on the key term under discussion.

    Many print and digital learning resources — adaptive or otherwise — are designed so that lessons basically involve students being shown how to execute a particular skill that they then practice. It sounds like Scott has seen the negative impact of this design in one adaptive technology program — where if a student gets a practice problem wrong, the only prescription available for support is more of the same direct explanation repeated again. More cowbell. More Neil Diamond.

    We need to evaluate new technology in light of learning principles and student performance just like we do with print resources. And we can’t lump all adaptive learning together just like we can’t overgeneralize about all classrooms, all worksheets, or all schools.

  16. > many of whom have never been in a class where math was anything but watching someone lecture about a procedure and then replicating that procedure twenty times on a piece of paper.

    Surely this is the real problem. Few teachers know or even attempt math as a social process.

    That’s why this is a valuable blog for both futurists and practitioners.

  17. Tim Hudson:

    We need to evaluate new technology in light of learning principles and student performance just like we do with print resources. And we can’t lump all adaptive learning together just like we can’t overgeneralize about all classrooms, all worksheets, or all schools.

    There are crucial differences between adaptive learning platforms, of course, but there are valid and useful ways to lump those platforms together.

    I’m happy to be corrected here but AFAIK no adaptive learning platform in 2014 can parse a student’s natural language conjecture and evaluate that conjecture as true or false.

    That generalization may not be true for all time but in 2014 it’s a fair generalization. And a crucial one, since constructing and critiquing conjectures is one of those eight big things students are supposed to be doing in a CCSS math class.

  18. Kevin Hall:

    Regarding the technical possibilities in 2014, this is an intelligent tutoring system that can interpret and respond to students’ spoken natural language dialogue.

    Sure. This kind of natural language parsing has been around since Eliza in the 1960s:

    If the student says something recognizable as the tutorial topic (e.g., “We made a circuit”), the system moves forward by asking the student what they know about the topic: You mentioned circuits. Can you tell me what a circuit is?

    Maybe I haven’t been clear about the kind of interaction I’m talking about, the kind that’s impossible for all adaptive learning systems in 2014 (AFAIK).

    T: “Pick any three consecutive whole numbers. Add them up. What do you notice?”
    S: “All our sums are divisible by three.”
    T: “Why do you think that is? Can you prove it’ll always happen?”
    S: “Easy. There are three numbers so they’re divisible by three.”

    A competent flesh-and-blood teacher knows what to do next and it isn’t to respond, “Tell me more about numbers that are divisible by three.”